# AI Simulations

Sixty-five interactive simulations of how modern AI works, each with a step-by-step code lab: a nine-step track on training a model from scratch, classical machine learning, neural network basics, real training runs, transformer internals, every major RAG design, vector search, evaluation, agents, guardrails and a large tabular foundation model lab. Most compute every number live in your browser, and many take your own input.

<div class="al al-sims" data-al><div class="al-bar"><div class="al-chips" role="group" aria-label="Filter by topic"><button type="button" data-al-tag="" aria-pressed="true">All <b>65</b></button><button type="button" data-al-tag="training from scratch" aria-pressed="false">Training from scratch <b>9</b></button><button type="button" data-al-tag="classical ml" aria-pressed="false">Classical ML <b>4</b></button><button type="button" data-al-tag="neural network basics" aria-pressed="false">Neural network basics <b>6</b></button><button type="button" data-al-tag="training" aria-pressed="false">Training <b>5</b></button><button type="button" data-al-tag="language models" aria-pressed="false">Language models <b>9</b></button><button type="button" data-al-tag="retrieval (rag)" aria-pressed="false">Retrieval (RAG) <b>15</b></button><button type="button" data-al-tag="agents" aria-pressed="false">Agents <b>6</b></button><button type="button" data-al-tag="large tabular foundation model" aria-pressed="false">Large tabular foundation model <b>11</b></button></div><label class="al-find"><span class="sr-only">Filter simulations by title</span><input type="search" placeholder="Filter by title" data-al-q autocomplete="off"></label></div><p class="al-none" hidden>No simulation matches. Try another word or topic.</p><div class="al-stats"><span><b>65</b> simulations</span><span><b>65</b> code labs</span><span><b>41</b> take your own input</span><span><b>60</b> compute every number live</span></div><nav class="al-path" aria-label="Learning path"><p>Start here: a path from one neuron to an evaluated agent</p><ol><li><a href="/ai-simulations/one-neuron"><span>01</span><b>One neuron</b><small>Neural network basics</small></a></li><li><a href="/ai-simulations/activation-functions"><span>02</span><b>Activations</b><small>Neural network basics</small></a></li><li><a href="/ai-simulations/loss-functions"><span>03</span><b>Loss functions</b><small>Neural network basics</small></a></li><li><a href="/ai-simulations/backpropagation"><span>04</span><b>Backpropagation</b><small>Neural network basics</small></a></li><li><a href="/ai-simulations/gradient-descent-optimisers"><span>05</span><b>Optimisers</b><small>Neural network basics</small></a></li><li><a href="/ai-simulations/training-run"><span>06</span><b>Training run</b><small>Training</small></a></li><li><a href="/ai-simulations/attention"><span>07</span><b>Attention</b><small>Language models</small></a></li><li><a href="/ai-simulations/rag-naive"><span>08</span><b>Naive RAG</b><small>Retrieval (RAG)</small></a></li><li><a href="/ai-simulations/rag-hybrid"><span>09</span><b>Hybrid search</b><small>Retrieval (RAG)</small></a></li><li><a href="/ai-simulations/react-tool-calling"><span>10</span><b>ReAct tool calling</b><small>Agents</small></a></li><li><a href="/ai-simulations/rag-evaluation-ragas"><span>11</span><b>RAG evaluation</b><small>Retrieval (RAG)</small></a></li></ol></nav><nav class="al-path al-path2" aria-label="Train a model from scratch"><p>Train a model from scratch, step by step: data to evaluation</p><ol><li><a href="/ai-simulations/train-data-pipeline"><span>01</span><b>Data pipeline</b><small>Training from scratch</small></a></li><li><a href="/ai-simulations/train-tokenizer-vocabulary"><span>02</span><b>Tokenizer and vocabulary</b><small>Training from scratch</small></a></li><li><a href="/ai-simulations/train-model-sizing"><span>03</span><b>Model sizing</b><small>Training from scratch</small></a></li><li><a href="/ai-simulations/train-scaling-laws"><span>04</span><b>Scaling laws</b><small>Training from scratch</small></a></li><li><a href="/ai-simulations/train-numeric-precision"><span>05</span><b>Numeric precision</b><small>Training from scratch</small></a></li><li><a href="/ai-simulations/train-parallelism-planner"><span>06</span><b>Parallelism planner</b><small>Training from scratch</small></a></li><li><a href="/ai-simulations/train-run-monitor"><span>07</span><b>Run, spikes and checkpoints</b><small>Training from scratch</small></a></li><li><a href="/ai-simulations/train-post-training-sft-dpo"><span>08</span><b>SFT and DPO</b><small>Training from scratch</small></a></li><li><a href="/ai-simulations/train-evaluation"><span>09</span><b>Evaluation</b><small>Training from scratch</small></a></li></ol></nav><a class="al-feat al-feat-sim" href="/ai-simulations/training-run" data-tag="training" data-q="a real training run, live: loss curves, checkpoints and a confusion matrix training run"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><path class="a-l2" d="M10.0 20.0L150.0 20.0"/><path class="a-l2" d="M10.0 38.0L150.0 38.0"/><path class="a-l2" d="M10.0 56.0L150.0 56.0"/><path class="a-l2" d="M10.0 74.0L150.0 74.0"/><path class="a-l" d="M12.0 22.0 L13.4 25.8 L14.8 26.7 L16.2 28.3 L17.6 32.2 L19.0 34.6 L20.4 34.8 L21.8 36.7 L23.2 40.3 L24.6 41.4 L26.0 41.3 L27.4 43.8 L28.8 46.6 L30.2 46.6 L31.6 46.8 L33.0 49.6 L34.4 51.3 L35.8 50.6 L37.2 51.5 L38.6 54.2 L40.0 54.8 L41.4 54.0 L42.8 55.5 L44.2 57.8 L45.6 57.4 L47.0 56.8 L48.4 58.9 L49.8 60.4 L51.2 59.3 L52.6 59.4 L54.0 61.7 L55.4 62.2 L56.8 60.9 L58.2 61.8 L59.6 63.9 L61.0 63.4 L62.4 62.3 L63.8 63.9 L65.2 65.4 L66.6 64.1 L68.0 63.7 L69.4 65.7 L70.8 66.3 L72.2 64.7 L73.6 65.2 L75.0 67.1 L76.4 66.7 L77.8 65.3 L79.2 66.6 L80.6 68.0 L82.0 66.8 L83.4 66.0 L84.8 67.8 L86.2 68.5 L87.6 66.8 L89.0 66.9 L90.4 68.8 L91.8 68.5 L93.2 67.0 L94.6 67.9 L96.0 69.5 L97.4 68.3 L98.8 67.3 L100.2 68.9 L101.6 69.7 L103.0 68.1 L104.4 67.8 L105.8 69.7 L107.2 69.6 L108.6 67.9 L110.0 68.6 L111.4 70.2 L112.8 69.3 L114.2 68.0 L115.6 69.4 L117.0 70.4 L118.4 68.8 L119.8 68.3 L121.2 70.1 L122.6 70.3 L124.0 68.5 L125.4 68.9 L126.8 70.6 L128.2 69.8 L129.6 68.4 L131.0 69.6 L132.4 70.8 L133.8 69.3 L135.2 68.5 L136.6 70.2 L138.0 70.6 L139.4 68.9 L140.8 69.0 L142.2 70.7 L143.6 70.2 L145.0 68.6 L146.4 69.6 L147.8 71.0 L149.2 69.7 L150.6 68.6"/><path class="a-b" d="M12.0 26.0 L26.0 44.8 L40.0 55.5 L54.0 61.7 L68.0 65.2 L82.0 67.3 L96.0 68.4 L110.0 69.1 L124.0 69.5 L138.0 69.7 L152.0 69.8"/><circle class="a-db" cx="12.0" cy="26.0" r="1.4"/><circle class="a-db" cx="26.0" cy="44.8" r="1.4"/><circle class="a-db" cx="40.0" cy="55.5" r="1.4"/><circle class="a-db" cx="54.0" cy="61.7" r="1.4"/><circle class="a-db" cx="68.0" cy="65.2" r="1.4"/><circle class="a-db" cx="82.0" cy="67.3" r="1.4"/><circle class="a-db" cx="96.0" cy="68.4" r="1.4"/><circle class="a-db" cx="110.0" cy="69.1" r="1.4"/><circle class="a-db" cx="124.0" cy="69.5" r="1.4"/><circle class="a-db" cx="138.0" cy="69.7" r="1.4"/><circle class="a-db" cx="152.0" cy="69.8" r="1.4"/></svg></span><span class="al-meta"><span class="al-tag">Training</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-lab">Code lab</span></span><span class="al-t">A real training run, live: loss curves, checkpoints and a confusion matrix</span><span class="al-d">Train a real neural network on 1,797 handwritten digits in your browser and watch it like a production run: loss, learning rate, checkpoints, errors.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><section class="al-month"><h2 class="al-mh" id="training-from-scratch">Training from scratch <span>9</span></h2><div class="al-grid"><a class="al-card" href="/ai-simulations/train-data-pipeline" data-tag="training from scratch" data-q="pre-training data: deduplicate, filter by quality and mix the sources data pipeline"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-g" x="16" y="22" width="16" height="52" rx="2"/><rect class="a-g" x="42" y="34" width="16" height="40" rx="2"/><rect class="a-g" x="68" y="43" width="16" height="31" rx="2"/><rect class="a-g" x="94" y="50" width="16" height="24" rx="2"/><rect class="a-g" x="120" y="54" width="16" height="20" rx="2"/></svg></span><span class="al-meta"><span class="al-tag">Training from scratch</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Pre-training data: deduplicate, filter by quality and mix the sources</span><span class="al-d">Clean a 3,000-document corpus live: remove exact and near duplicates, filter by quality and language, then weight the sources into a training mixture.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/train-tokenizer-vocabulary" data-tag="training from scratch" data-q="train a tokenizer: byte-pair merges and the vocabulary-size trade-off tokenizer and vocabulary"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><path class="a-l2" d="M12 74L150 74"/><path class="a-b" d="M12.0 72.0 L14.8 66.5 L17.6 61.6 L20.4 57.3 L23.2 53.3 L26.0 49.8 L28.8 46.7 L31.6 43.9 L34.4 41.4 L37.2 39.1 L40.0 37.1 L42.8 35.3 L45.6 33.7 L48.4 32.3 L51.2 31.0 L54.0 29.8 L56.8 28.8 L59.6 27.9 L62.4 27.0 L65.2 26.3 L68.0 25.6 L70.8 25.0 L73.6 24.5 L76.4 24.0 L79.2 23.6 L82.0 23.2 L84.8 22.9 L87.6 22.6 L90.4 22.3 L93.2 22.1 L96.0 21.9 L98.8 21.7 L101.6 21.5 L104.4 21.3 L107.2 21.2 L110.0 21.1 L112.8 21.0 L115.6 20.9 L118.4 20.8 L121.2 20.7 L124.0 20.6 L126.8 20.5 L129.6 20.5 L132.4 20.4 L135.2 20.4 L138.0 20.4 L140.8 20.3 L143.6 20.3 L146.4 20.3 L149.2 20.2"/></svg></span><span class="al-meta"><span class="al-tag">Training from scratch</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Train a tokenizer: byte-pair merges and the vocabulary-size trade-off</span><span class="al-d">Learn byte-pair merges live, tokenize your own text with them and see how vocabulary size trades sequence length against embedding parameters.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/train-model-sizing" data-tag="training from scratch" data-q="size the model: parameters, flops, kv cache and training memory model sizing"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-g" x="18" y="14" width="12" height="12" rx="2"/><rect class="a-g" x="34" y="14" width="12" height="12" rx="2"/><rect class="a-g" x="50" y="14" width="12" height="12" rx="2"/><rect class="a-g" x="66" y="14" width="12" height="12" rx="2"/><rect class="a-g" x="82" y="14" width="12" height="12" rx="2"/><rect class="a-g" x="98" y="14" width="12" height="12" rx="2"/><rect class="a-g" x="114" y="14" width="12" height="12" rx="2"/><rect class="a-g" x="130" y="14" width="12" height="12" rx="2"/><rect class="a-g" x="18" y="30" width="12" height="12" rx="2"/><rect class="a-g" x="34" y="30" width="12" height="12" rx="2"/><rect class="a-g" x="50" y="30" width="12" height="12" rx="2"/><rect class="a-g" x="66" y="30" width="12" height="12" rx="2"/><rect class="a-g" x="82" y="30" width="12" height="12" rx="2"/><rect class="a-g" x="98" y="30" width="12" height="12" rx="2"/><rect class="a-g" x="114" y="30" width="12" height="12" rx="2"/><rect class="a-g" x="130" y="30" width="12" height="12" rx="2"/><rect class="a-g" x="18" y="46" width="12" height="12" rx="2"/><rect class="a-g" x="34" y="46" width="12" height="12" rx="2"/><rect class="a-g" x="50" y="46" width="12" height="12" rx="2"/><rect class="a-g" x="66" y="46" width="12" height="12" rx="2"/><rect class="a-g" x="82" y="46" width="12" height="12" rx="2"/><rect class="a-g" x="98" y="46" width="12" height="12" rx="2"/><rect class="a-g" x="114" y="46" width="12" height="12" rx="2"/><rect class="a-g" x="130" y="46" width="12" height="12" rx="2"/><rect class="a-g" x="18" y="62" width="12" height="12" rx="2"/><rect class="a-g" x="34" y="62" width="12" height="12" rx="2"/><rect class="a-g" x="50" y="62" width="12" height="12" rx="2"/><rect class="a-g" x="66" y="62" width="12" height="12" rx="2"/><rect class="a-g" x="82" y="62" width="12" height="12" rx="2"/><rect class="a-g" x="98" y="62" width="12" height="12" rx="2"/><rect class="a-g" x="114" y="62" width="12" height="12" rx="2"/><rect class="a-g" x="130" y="62" width="12" height="12" rx="2"/></svg></span><span class="al-meta"><span class="al-tag">Training from scratch</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Size the model: parameters, FLOPs, KV cache and training memory</span><span class="al-d">Set layers, width, heads and vocabulary and get the parameter count, training FLOPs per token, KV cache and memory per GPU, from the standard formulas.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/train-scaling-laws" data-tag="training from scratch" data-q="scaling laws: how to split a compute budget between parameters and tokens scaling laws"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><path class="a-l2" d="M12 74L150 74"/><path class="a-b" d="M12.0 53.7 L14.8 51.1 L17.6 48.6 L20.4 46.3 L23.2 44.0 L26.0 41.9 L28.8 39.8 L31.6 37.9 L34.4 36.1 L37.2 34.4 L40.0 32.8 L42.8 31.3 L45.6 29.9 L48.4 28.7 L51.2 27.5 L54.0 26.5 L56.8 25.5 L59.6 24.7 L62.4 24.0 L65.2 23.4 L68.0 22.9 L70.8 22.5 L73.6 22.2 L76.4 22.1 L79.2 22.0 L82.0 22.1 L84.8 22.2 L87.6 22.5 L90.4 22.9 L93.2 23.4 L96.0 24.0 L98.8 24.7 L101.6 25.5 L104.4 26.5 L107.2 27.5 L110.0 28.7 L112.8 29.9 L115.6 31.3 L118.4 32.8 L121.2 34.4 L124.0 36.1 L126.8 37.9 L129.6 39.8 L132.4 41.9 L135.2 44.0 L138.0 46.3 L140.8 48.6 L143.6 51.1 L146.4 53.7 L149.2 56.4"/><circle class="a-d" cx="79" cy="22" r="3"/></svg></span><span class="al-meta"><span class="al-tag">Training from scratch</span><span>Advanced</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Scaling laws: how to split a compute budget between parameters and tokens</span><span class="al-d">Spend one compute budget on a bigger model or on more tokens, with the Chinchilla loss law, and see why serving costs push models to train past it.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/train-numeric-precision" data-tag="training from scratch" data-q="numeric precision: fp32, bf16, fp16 and fp8 in training, bit by bit numeric precision"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-g" x="14" y="34" width="7" height="16" rx="1"/><rect class="a-b" x="22" y="34" width="7" height="16" rx="1"/><rect class="a-b" x="31" y="34" width="7" height="16" rx="1"/><rect class="a-b" x="39" y="34" width="7" height="16" rx="1"/><rect class="a-b" x="48" y="34" width="7" height="16" rx="1"/><rect class="a-b" x="56" y="34" width="7" height="16" rx="1"/><rect class="a-b" x="65" y="34" width="7" height="16" rx="1"/><rect class="a-b" x="73" y="34" width="7" height="16" rx="1"/><rect class="a-b" x="82" y="34" width="7" height="16" rx="1"/><rect class="a-g" x="90" y="34" width="7" height="16" rx="1"/><rect class="a-g" x="99" y="34" width="7" height="16" rx="1"/><rect class="a-g" x="107" y="34" width="7" height="16" rx="1"/><rect class="a-g" x="116" y="34" width="7" height="16" rx="1"/><rect class="a-g" x="124" y="34" width="7" height="16" rx="1"/><rect class="a-g" x="133" y="34" width="7" height="16" rx="1"/><rect class="a-g" x="141" y="34" width="7" height="16" rx="1"/></svg></span><span class="al-meta"><span class="al-tag">Training from scratch</span><span>Advanced</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Numeric precision: fp32, bf16, fp16 and fp8 in training, bit by bit</span><span class="al-d">Type a number and see it stored in fp32, bf16, fp16 and fp8, then watch small gradients vanish in fp16 and come back with loss scaling.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/train-parallelism-planner" data-tag="training from scratch" data-q="parallelism planner: fit a run on gpus with data, tensor and pipeline parallelism parallelism planner"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-g" x="18" y="14" width="12" height="12" rx="2"/><rect class="a-g" x="34" y="14" width="12" height="12" rx="2"/><rect class="a-g" x="50" y="14" width="12" height="12" rx="2"/><rect class="a-g" x="66" y="14" width="12" height="12" rx="2"/><rect class="a-g" x="82" y="14" width="12" height="12" rx="2"/><rect class="a-g" x="98" y="14" width="12" height="12" rx="2"/><rect class="a-g" x="114" y="14" width="12" height="12" rx="2"/><rect class="a-g" x="130" y="14" width="12" height="12" rx="2"/><rect class="a-g" x="18" y="30" width="12" height="12" rx="2"/><rect class="a-g" x="34" y="30" width="12" height="12" rx="2"/><rect class="a-g" x="50" y="30" width="12" height="12" rx="2"/><rect class="a-g" x="66" y="30" width="12" height="12" rx="2"/><rect class="a-g" x="82" y="30" width="12" height="12" rx="2"/><rect class="a-g" x="98" y="30" width="12" height="12" rx="2"/><rect class="a-g" x="114" y="30" width="12" height="12" rx="2"/><rect class="a-g" x="130" y="30" width="12" height="12" rx="2"/><rect class="a-g" x="18" y="46" width="12" height="12" rx="2"/><rect class="a-g" x="34" y="46" width="12" height="12" rx="2"/><rect class="a-g" x="50" y="46" width="12" height="12" rx="2"/><rect class="a-g" x="66" y="46" width="12" height="12" rx="2"/><rect class="a-g" x="82" y="46" width="12" height="12" rx="2"/><rect class="a-g" x="98" y="46" width="12" height="12" rx="2"/><rect class="a-g" x="114" y="46" width="12" height="12" rx="2"/><rect class="a-g" x="130" y="46" width="12" height="12" rx="2"/><rect class="a-g" x="18" y="62" width="12" height="12" rx="2"/><rect class="a-g" x="34" y="62" width="12" height="12" rx="2"/><rect class="a-g" x="50" y="62" width="12" height="12" rx="2"/><rect class="a-g" x="66" y="62" width="12" height="12" rx="2"/><rect class="a-g" x="82" y="62" width="12" height="12" rx="2"/><rect class="a-g" x="98" y="62" width="12" height="12" rx="2"/><rect class="a-g" x="114" y="62" width="12" height="12" rx="2"/><rect class="a-g" x="130" y="62" width="12" height="12" rx="2"/></svg></span><span class="al-meta"><span class="al-tag">Training from scratch</span><span>Advanced</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Parallelism planner: fit a run on GPUs with data, tensor and pipeline parallelism</span><span class="al-d">Split a 1B to 405B model across up to 4,096 GPUs with data, tensor and pipeline parallelism and ZeRO, and see memory per GPU and the pipeline bubble.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/train-run-monitor" data-tag="training from scratch" data-q="a training run, live: warmup, gradient clipping, loss spikes and checkpoints run, spikes and checkpoints"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><path class="a-l2" d="M12 74L150 74"/><path class="a-b" d="M12.0 72.0 L14.8 66.5 L17.6 61.6 L20.4 57.3 L23.2 53.3 L26.0 49.8 L28.8 46.7 L31.6 43.9 L34.4 41.4 L37.2 39.1 L40.0 37.1 L42.8 35.3 L45.6 33.7 L48.4 32.3 L51.2 31.0 L54.0 29.8 L56.8 28.8 L59.6 27.9 L62.4 27.0 L65.2 26.3 L68.0 25.6 L70.8 25.0 L73.6 24.5 L76.4 24.0 L79.2 23.6 L82.0 23.2 L84.8 22.9 L87.6 22.6 L90.4 22.3 L93.2 22.1 L96.0 21.9 L98.8 21.7 L101.6 21.5 L104.4 21.3 L107.2 21.2 L110.0 21.1 L112.8 21.0 L115.6 20.9 L118.4 20.8 L121.2 20.7 L124.0 20.6 L126.8 20.5 L129.6 20.5 L132.4 20.4 L135.2 20.4 L138.0 20.4 L140.8 20.3 L143.6 20.3 L146.4 20.3 L149.2 20.2"/></svg></span><span class="al-meta"><span class="al-tag">Training from scratch</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">A training run, live: warmup, gradient clipping, loss spikes and checkpoints</span><span class="al-d">Train a real network in your browser with AdamW and warmup, inject a bad batch, watch clipping absorb it and roll back to a checkpoint.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/train-post-training-sft-dpo" data-tag="training from scratch" data-q="post-training: supervised fine-tuning, then preference optimisation with dpo sft and dpo"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-g" x="18" y="18" width="56" height="16" rx="3"/><rect class="a-g" x="86" y="18" width="56" height="16" rx="3"/><path class="a-b" d="M46 42 L46 60 L114 60 L114 42"/><circle class="a-d" cx="46" cy="70" r="3"/></svg></span><span class="al-meta"><span class="al-tag">Training from scratch</span><span>Advanced</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Post-training: supervised fine-tuning, then preference optimisation with DPO</span><span class="al-d">Fine-tune a toy policy on demonstrations, then align it with DPO on preference pairs, and see how mislabelled preferences teach the wrong lesson.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/train-evaluation" data-tag="training from scratch" data-q="evaluation: confidence intervals, leaked test data and pass@k evaluation"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-g" x="16" y="22" width="16" height="52" rx="2"/><rect class="a-g" x="42" y="34" width="16" height="40" rx="2"/><rect class="a-g" x="68" y="43" width="16" height="31" rx="2"/><rect class="a-g" x="94" y="50" width="16" height="24" rx="2"/><rect class="a-g" x="120" y="54" width="16" height="20" rx="2"/></svg></span><span class="al-meta"><span class="al-tag">Training from scratch</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Evaluation: confidence intervals, leaked test data and pass@k</span><span class="al-d">See how much a benchmark score moves between runs, when a two-point gap is real, how leaked test items inflate a score and how pass@k works.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a></div></section><section class="al-month"><h2 class="al-mh" id="classical-ml">Classical ML <span>4</span></h2><div class="al-grid"><a class="al-card" href="/ai-simulations/k-means-clustering" data-tag="classical ml" data-q="k-means clustering, step by step k-means"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><circle class="a-d" cx="54.0" cy="37.1" r="0.9"/><circle class="a-d" cx="45.5" cy="26.3" r="0.9"/><circle class="a-d" cx="37.4" cy="30.2" r="0.9"/><circle class="a-d" cx="37.8" cy="23.0" r="0.9"/><circle class="a-d" cx="46.4" cy="30.7" r="0.9"/><circle class="a-d" cx="48.8" cy="25.5" r="0.9"/><circle class="a-d" cx="45.0" cy="29.7" r="0.9"/><circle class="a-d" cx="34.5" cy="32.6" r="0.9"/><circle class="a-d" cx="47.2" cy="41.7" r="0.9"/><circle class="a-d" cx="46.4" cy="29.3" r="0.9"/><circle class="a-d" cx="53.6" cy="31.0" r="0.9"/><circle class="a-d" cx="51.4" cy="28.2" r="0.9"/><circle class="a-d" cx="46.5" cy="35.0" r="0.9"/><circle class="a-d" cx="49.9" cy="30.6" r="0.9"/><circle class="a-d" cx="37.4" cy="32.2" r="0.9"/><circle class="a-d" cx="45.5" cy="33.5" r="0.9"/><circle class="a-d" cx="46.5" cy="35.3" r="0.9"/><circle class="a-d" cx="44.6" cy="31.0" r="0.9"/><circle class="a-d" cx="49.7" cy="24.7" r="0.9"/><circle class="a-d" cx="42.2" cy="27.5" r="0.9"/><circle class="a-d" cx="58.9" cy="29.5" r="0.9"/><circle class="a-d" cx="49.6" cy="33.0" r="0.9"/><circle class="a-d" cx="43.0" cy="22.4" r="0.9"/><circle class="a-d" cx="51.8" cy="28.0" r="0.9"/><circle class="a-d" cx="50.0" cy="23.6" r="0.9"/><circle class="a-d" cx="41.9" cy="36.2" r="0.9"/><circle class="a-d" cx="55.0" cy="23.6" r="0.9"/><circle class="a-d" cx="35.7" cy="29.8" r="0.9"/><circle class="a-d" cx="50.1" cy="30.8" r="0.9"/><circle class="a-d" cx="47.1" cy="25.2" r="0.9"/><circle class="a-d" cx="49.1" cy="35.5" r="0.9"/><circle class="a-d" cx="42.0" cy="23.0" r="0.9"/><circle class="a-d" cx="39.7" cy="33.7" r="0.9"/><circle class="a-d" cx="32.9" cy="29.5" r="0.9"/><circle class="a-d" cx="38.1" cy="29.4" r="0.9"/><circle class="a-d" cx="43.3" cy="30.1" r="0.9"/><circle class="a-d" cx="55.5" cy="32.1" r="0.9"/><circle class="a-d" cx="54.3" cy="29.3" r="0.9"/><circle class="a-d" cx="41.6" cy="31.9" r="0.9"/><circle class="a-d" cx="25.1" cy="29.8" r="0.9"/><circle class="a-d" cx="128.7" cy="28.3" r="0.9"/><circle class="a-d" cx="113.2" cy="32.8" r="0.9"/><circle class="a-d" cx="116.7" cy="24.1" r="0.9"/><circle class="a-d" cx="106.7" cy="27.8" r="0.9"/><circle class="a-d" cx="101.4" cy="27.3" r="0.9"/><circle class="a-d" cx="105.9" cy="30.4" r="0.9"/><circle class="a-d" cx="102.7" cy="34.4" r="0.9"/><circle class="a-d" cx="105.6" cy="14.1" r="0.9"/><circle class="a-d" cx="119.5" cy="29.8" r="0.9"/><circle class="a-d" cx="104.1" cy="33.5" r="0.9"/><circle class="a-d" cx="111.8" cy="32.3" r="0.9"/><circle class="a-d" cx="103.2" cy="33.1" r="0.9"/><circle class="a-d" cx="97.7" cy="40.1" r="0.9"/><circle class="a-d" cx="99.9" cy="30.8" r="0.9"/><circle class="a-d" cx="110.2" cy="33.2" r="0.9"/><circle class="a-d" cx="108.0" cy="34.7" r="0.9"/><circle class="a-d" cx="80.9" cy="30.7" r="0.9"/><circle class="a-d" cx="107.7" cy="28.8" r="0.9"/><circle class="a-d" cx="121.2" cy="25.8" r="0.9"/><circle class="a-d" cx="108.3" cy="19.9" r="0.9"/><circle class="a-d" cx="111.1" cy="22.1" r="0.9"/><circle class="a-d" cx="96.3" cy="44.5" r="0.9"/><circle class="a-d" cx="114.6" cy="31.2" r="0.9"/><circle class="a-d" cx="110.3" cy="23.1" r="0.9"/><circle class="a-d" cx="100.5" cy="33.7" r="0.9"/><circle class="a-d" cx="91.8" cy="32.8" r="0.9"/><circle class="a-d" cx="94.9" cy="31.9" r="0.9"/><circle class="a-d" cx="99.9" cy="41.2" r="0.9"/><circle class="a-d" cx="117.2" cy="28.3" r="0.9"/><circle class="a-d" cx="93.6" cy="26.8" r="0.9"/><circle class="a-d" cx="108.5" cy="25.6" r="0.9"/><circle class="a-d" cx="111.2" cy="36.9" r="0.9"/><circle class="a-d" cx="108.5" cy="28.8" r="0.9"/><circle class="a-d" cx="115.3" cy="29.6" r="0.9"/><circle class="a-d" cx="115.8" cy="29.5" r="0.9"/><circle class="a-d" cx="122.1" cy="29.7" r="0.9"/><circle class="a-d" cx="100.3" cy="31.8" r="0.9"/><circle class="a-d" cx="103.8" cy="25.9" r="0.9"/><circle class="a-d" cx="107.9" cy="35.5" r="0.9"/><circle class="a-d" cx="91.4" cy="31.0" r="0.9"/><circle class="a-d" cx="80.7" cy="72.1" r="0.9"/><circle class="a-d" cx="73.5" cy="70.9" r="0.9"/><circle class="a-d" cx="78.2" cy="64.7" r="0.9"/><circle class="a-d" cx="93.3" cy="66.8" r="0.9"/><circle class="a-d" cx="79.7" cy="69.6" r="0.9"/><circle class="a-d" cx="87.9" cy="65.8" r="0.9"/><circle class="a-d" cx="84.1" cy="61.2" r="0.9"/><circle class="a-d" cx="77.4" cy="63.9" r="0.9"/><circle class="a-d" cx="70.7" cy="58.6" r="0.9"/><circle class="a-d" cx="68.6" cy="64.8" r="0.9"/><circle class="a-d" cx="78.8" cy="64.4" r="0.9"/><circle class="a-d" cx="80.5" cy="59.5" r="0.9"/><circle class="a-d" cx="79.4" cy="67.2" r="0.9"/><circle class="a-d" cx="85.3" cy="61.9" r="0.9"/><circle class="a-d" cx="77.2" cy="56.1" r="0.9"/><circle class="a-d" cx="76.5" cy="55.2" r="0.9"/><circle class="a-d" cx="70.1" cy="71.4" r="0.9"/><circle class="a-d" cx="64.6" cy="69.9" r="0.9"/><circle class="a-d" cx="82.3" cy="64.5" r="0.9"/><circle class="a-d" cx="83.2" cy="68.6" r="0.9"/><circle class="a-d" cx="87.3" cy="64.9" r="0.9"/><circle class="a-d" cx="75.9" cy="63.0" r="0.9"/><circle class="a-d" cx="73.1" cy="65.8" r="0.9"/><circle class="a-d" cx="74.5" cy="71.2" r="0.9"/><circle class="a-d" cx="66.9" cy="60.6" r="0.9"/><circle class="a-d" cx="73.3" cy="55.7" r="0.9"/><circle class="a-d" cx="93.3" cy="54.2" r="0.9"/><circle class="a-d" cx="78.0" cy="63.4" r="0.9"/><circle class="a-d" cx="91.6" cy="56.3" r="0.9"/><circle class="a-d" cx="87.5" cy="62.4" r="0.9"/><circle class="a-d" cx="78.9" cy="62.7" r="0.9"/><circle class="a-d" cx="84.5" cy="60.4" r="0.9"/><circle class="a-d" cx="79.4" cy="67.7" r="0.9"/><circle class="a-d" cx="92.9" cy="54.2" r="0.9"/><circle class="a-d" cx="90.7" cy="70.6" r="0.9"/><circle class="a-d" cx="76.6" cy="67.5" r="0.9"/><circle class="a-d" cx="76.7" cy="74.1" r="0.9"/><circle class="a-d" cx="81.5" cy="64.9" r="0.9"/><circle class="a-d" cx="78.4" cy="65.0" r="0.9"/><circle class="a-d" cx="78.8" cy="61.7" r="0.9"/><circle class="a-db" cx="45.0" cy="30.0" r="2.6"/><circle class="a-db" cx="110.0" cy="32.0" r="2.6"/><circle class="a-db" cx="80.0" cy="66.0" r="2.6"/></svg></span><span class="al-meta"><span class="al-tag">Classical ML</span><span>Beginner</span><span class="al-real">Real computation</span><span class="al-lab">Code lab</span></span><span class="al-t">k-means clustering, step by step</span><span class="al-d">Watch k-means group unlabeled points: assign each point to its nearest centre, move each centre to the mean, repeat until nothing changes.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/linear-regression" data-tag="classical ml" data-q="linear regression: your points, gradient descent and the loss surface linear regression"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><path class="a-l2" d="M10.0 20.0L150.0 20.0"/><path class="a-l2" d="M10.0 38.0L150.0 38.0"/><path class="a-l2" d="M10.0 56.0L150.0 56.0"/><path class="a-l2" d="M10.0 74.0L150.0 74.0"/><circle class="a-d" cx="16.0" cy="74.7" r="1.3"/><circle class="a-d" cx="24.4" cy="73.1" r="1.3"/><circle class="a-d" cx="32.8" cy="75.1" r="1.3"/><circle class="a-d" cx="41.2" cy="60.6" r="1.3"/><circle class="a-d" cx="49.6" cy="57.0" r="1.3"/><circle class="a-d" cx="58.0" cy="47.6" r="1.3"/><circle class="a-d" cx="66.4" cy="52.7" r="1.3"/><circle class="a-d" cx="74.8" cy="45.6" r="1.3"/><circle class="a-d" cx="83.2" cy="45.5" r="1.3"/><circle class="a-d" cx="91.6" cy="35.6" r="1.3"/><circle class="a-d" cx="100.0" cy="32.2" r="1.3"/><circle class="a-d" cx="108.4" cy="27.6" r="1.3"/><circle class="a-d" cx="116.8" cy="23.2" r="1.3"/><circle class="a-d" cx="125.2" cy="26.7" r="1.3"/><circle class="a-d" cx="133.6" cy="29.9" r="1.3"/><circle class="a-d" cx="142.0" cy="21.2" r="1.3"/><path class="a-b" d="M12.0 74.0L148.0 18.0"/><path class="a-l" d="M12.0 70.0L148.0 22.0"/></svg></span><span class="al-meta"><span class="al-tag">Classical ML</span><span>Beginner</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Linear regression: your points, gradient descent and the loss surface</span><span class="al-d">Click to place points and watch gradient descent fit a line or curve, with the loss surface, the exact least-squares answer, R² and MSE, MAE or Huber.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/decision-tree" data-tag="classical ml" data-q="decision trees: how cart chooses every split decision tree"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-g" x="12.0" y="10.0" width="136.0" height="70.0" rx="2"/><path class="a-b" d="M70.0 10.0L70.0 80.0"/><path class="a-l" d="M12.0 44.0L70.0 44.0"/><path class="a-l" d="M70.0 58.0L148.0 58.0"/><path class="a-l2" d="M110.0 10.0L110.0 58.0"/><rect class="a-f" x="12.0" y="44.0" width="58.0" height="36.0" rx="0"/><rect class="a-f" x="110.0" y="10.0" width="38.0" height="48.0" rx="0"/><circle class="a-d" cx="45.5" cy="32.5" r="0.9"/><circle class="a-d" cx="34.7" cy="21.8" r="0.9"/><circle class="a-d" cx="42.6" cy="34.5" r="0.9"/><circle class="a-d" cx="43.0" cy="20.5" r="0.9"/><circle class="a-d" cx="34.9" cy="25.9" r="0.9"/><circle class="a-d" cx="46.7" cy="26.1" r="0.9"/><circle class="a-d" cx="37.7" cy="25.1" r="0.9"/><circle class="a-d" cx="41.4" cy="22.6" r="0.9"/><circle class="a-d" cx="40.2" cy="27.4" r="0.9"/><circle class="a-d" cx="51.5" cy="25.9" r="0.9"/><circle class="a-db" cx="103.2" cy="69.4" r="0.9"/><circle class="a-db" cx="121.5" cy="72.3" r="0.9"/><circle class="a-db" cx="101.2" cy="72.0" r="0.9"/><circle class="a-db" cx="96.8" cy="66.9" r="0.9"/><circle class="a-db" cx="103.3" cy="68.0" r="0.9"/><circle class="a-db" cx="105.1" cy="69.1" r="0.9"/><circle class="a-db" cx="88.8" cy="64.2" r="0.9"/><circle class="a-db" cx="94.6" cy="61.2" r="0.9"/><circle class="a-db" cx="99.5" cy="68.7" r="0.9"/><circle class="a-db" cx="105.9" cy="66.2" r="0.9"/><circle class="a-db" cx="53.0" cy="57.3" r="0.9"/><circle class="a-db" cx="42.0" cy="60.9" r="0.9"/><circle class="a-db" cx="34.7" cy="59.6" r="0.9"/><circle class="a-db" cx="36.9" cy="61.6" r="0.9"/><circle class="a-db" cx="42.3" cy="66.3" r="0.9"/><circle class="a-db" cx="41.2" cy="66.6" r="0.9"/><circle class="a-db" cx="42.5" cy="62.3" r="0.9"/><circle class="a-db" cx="38.4" cy="57.7" r="0.9"/><circle class="a-db" cx="127.3" cy="24.2" r="0.9"/><circle class="a-db" cx="138.8" cy="45.7" r="0.9"/><circle class="a-db" cx="133.1" cy="37.9" r="0.9"/><circle class="a-db" cx="138.5" cy="35.9" r="0.9"/><circle class="a-db" cx="135.5" cy="37.4" r="0.9"/><circle class="a-db" cx="128.4" cy="24.9" r="0.9"/><circle class="a-db" cx="118.8" cy="35.0" r="0.9"/><circle class="a-db" cx="133.0" cy="31.8" r="0.9"/></svg></span><span class="al-meta"><span class="al-tag">Classical ML</span><span>Beginner</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Decision trees: how CART chooses every split</span><span class="al-d">Watch a decision tree split your points with Gini impurity or entropy, change depth and leaf size, and see when it starts carving out noise.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/pca" data-tag="classical ml" data-q="pca: principal components from the covariance matrix pca"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><ellipse class="a-l" cx="80" cy="45" rx="58" ry="16" transform="rotate(-28 80 45)"/><circle class="a-d" cx="97.3" cy="30.6" r="1.0"/><circle class="a-d" cx="106.7" cy="31.1" r="1.0"/><circle class="a-d" cx="129.2" cy="21.0" r="1.0"/><circle class="a-d" cx="77.7" cy="35.9" r="1.0"/><circle class="a-d" cx="82.1" cy="47.0" r="1.0"/><circle class="a-d" cx="61.4" cy="68.7" r="1.0"/><circle class="a-d" cx="93.8" cy="46.0" r="1.0"/><circle class="a-d" cx="78.3" cy="54.5" r="1.0"/><circle class="a-d" cx="86.1" cy="37.8" r="1.0"/><circle class="a-d" cx="61.6" cy="57.3" r="1.0"/><circle class="a-d" cx="59.0" cy="53.1" r="1.0"/><circle class="a-d" cx="51.2" cy="51.9" r="1.0"/><circle class="a-d" cx="51.7" cy="63.8" r="1.0"/><circle class="a-d" cx="76.1" cy="41.9" r="1.0"/><circle class="a-d" cx="101.3" cy="24.4" r="1.0"/><circle class="a-d" cx="81.1" cy="39.7" r="1.0"/><circle class="a-d" cx="67.4" cy="58.1" r="1.0"/><circle class="a-d" cx="47.1" cy="72.6" r="1.0"/><circle class="a-d" cx="92.4" cy="37.6" r="1.0"/><circle class="a-d" cx="63.5" cy="52.6" r="1.0"/><circle class="a-d" cx="108.0" cy="34.7" r="1.0"/><circle class="a-d" cx="94.4" cy="33.2" r="1.0"/><circle class="a-d" cx="94.3" cy="45.4" r="1.0"/><circle class="a-d" cx="124.2" cy="20.0" r="1.0"/><circle class="a-d" cx="72.8" cy="53.9" r="1.0"/><circle class="a-d" cx="43.4" cy="57.8" r="1.0"/><circle class="a-d" cx="75.1" cy="36.4" r="1.0"/><circle class="a-d" cx="61.6" cy="56.6" r="1.0"/><circle class="a-d" cx="88.8" cy="41.5" r="1.0"/><circle class="a-d" cx="63.4" cy="57.6" r="1.0"/><circle class="a-d" cx="63.0" cy="53.0" r="1.0"/><circle class="a-d" cx="87.0" cy="34.0" r="1.0"/><circle class="a-d" cx="94.6" cy="41.9" r="1.0"/><circle class="a-d" cx="48.2" cy="48.9" r="1.0"/><circle class="a-d" cx="52.3" cy="58.1" r="1.0"/><circle class="a-d" cx="48.4" cy="64.7" r="1.0"/><circle class="a-d" cx="52.2" cy="60.3" r="1.0"/><circle class="a-d" cx="77.3" cy="49.0" r="1.0"/><circle class="a-d" cx="64.6" cy="49.8" r="1.0"/><circle class="a-d" cx="88.4" cy="48.4" r="1.0"/><circle class="a-d" cx="39.4" cy="67.5" r="1.0"/><circle class="a-d" cx="70.5" cy="53.6" r="1.0"/><circle class="a-d" cx="73.9" cy="40.7" r="1.0"/><circle class="a-d" cx="113.7" cy="33.6" r="1.0"/><circle class="a-d" cx="57.0" cy="54.0" r="1.0"/><circle class="a-d" cx="76.9" cy="47.1" r="1.0"/><circle class="a-d" cx="91.8" cy="28.8" r="1.0"/><circle class="a-d" cx="68.6" cy="55.2" r="1.0"/><circle class="a-d" cx="66.3" cy="44.6" r="1.0"/><circle class="a-d" cx="98.0" cy="31.8" r="1.0"/><circle class="a-d" cx="65.6" cy="57.1" r="1.0"/><circle class="a-d" cx="79.4" cy="41.5" r="1.0"/><circle class="a-d" cx="89.1" cy="45.6" r="1.0"/><circle class="a-d" cx="50.9" cy="67.8" r="1.0"/><circle class="a-d" cx="95.0" cy="50.0" r="1.0"/><circle class="a-d" cx="124.5" cy="25.7" r="1.0"/><circle class="a-d" cx="104.4" cy="28.5" r="1.0"/><circle class="a-d" cx="106.7" cy="31.0" r="1.0"/><circle class="a-d" cx="71.2" cy="58.0" r="1.0"/><circle class="a-d" cx="87.2" cy="39.8" r="1.0"/><path class="a-b" d="M25.3 74.2L134.7 15.8"/><path class="a-l" d="M71.5 29.1L88.5 60.9"/></svg></span><span class="al-meta"><span class="al-tag">Classical ML</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">PCA: principal components from the covariance matrix</span><span class="al-d">See principal component analysis compute the covariance matrix, its eigenvectors and the variance each component keeps, and project your points onto PC1.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a></div></section><section class="al-month"><h2 class="al-mh" id="neural-network-basics">Neural network basics <span>6</span></h2><div class="al-grid"><a class="al-card" href="/ai-simulations/one-neuron" data-tag="neural network basics" data-q="one neuron: w·x + b and an activation one neuron"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><polygon class="a-f" points="52,84 128,6 150,6 150,84"/><path class="a-b" d="M52.0 84.0L128.0 6.0"/><path class="a-l" d="M90.0 45.0L104.0 56.0"/><circle class="a-db" cx="104.0" cy="56.0" r="1.8"/><circle class="a-d" cx="70.0" cy="40.0" r="2.2"/></svg></span><span class="al-meta"><span class="al-tag">Neural network basics</span><span>Beginner</span><span class="al-real">Real computation</span><span class="al-lab">Code lab</span></span><span class="al-t">One neuron: w·x + b and an activation</span><span class="al-d">Move the weights and bias of a single neuron and watch its decision line turn and shift, with every number of w·x + b computed live.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/activation-functions" data-tag="neural network basics" data-q="activation functions: relu, gelu, sigmoid, tanh and more activations"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><path class="a-l2" d="M10.0 60.0L150.0 60.0"/><path class="a-l2" d="M80.0 8.0L80.0 84.0"/><path class="a-l" d="M10.0 60.0 L11.0 60.0 L12.0 60.0 L13.0 60.0 L14.0 60.0 L15.0 60.0 L16.0 60.0 L17.0 59.9 L18.0 59.9 L19.0 59.9 L20.0 59.9 L21.0 59.9 L22.0 59.9 L23.0 59.9 L24.0 59.9 L25.0 59.9 L26.0 59.9 L27.0 59.9 L28.0 59.8 L29.0 59.8 L30.0 59.8 L31.0 59.8 L32.0 59.8 L33.0 59.7 L34.0 59.7 L35.0 59.7 L36.0 59.6 L37.0 59.6 L38.0 59.6 L39.0 59.5 L40.0 59.5 L41.0 59.4 L42.0 59.3 L43.0 59.3 L44.0 59.2 L45.0 59.1 L46.0 59.0 L47.0 58.9 L48.0 58.8 L49.0 58.7 L50.0 58.6 L51.0 58.4 L52.0 58.3 L53.0 58.1 L54.0 57.9 L55.0 57.7 L56.0 57.5 L57.0 57.3 L58.0 57.0 L59.0 56.7 L60.0 56.4 L61.0 56.1 L62.0 55.7 L63.0 55.4 L64.0 55.0 L65.0 54.5 L66.0 54.1 L67.0 53.6 L68.0 53.1 L69.0 52.5 L70.0 51.9 L71.0 51.3 L72.0 50.7 L73.0 50.0 L74.0 49.4 L75.0 48.7 L76.0 48.0 L77.0 47.2 L78.0 46.5 L79.0 45.7 L80.0 45.0 L81.0 44.3 L82.0 43.5 L83.0 42.8 L84.0 42.0 L85.0 41.3 L86.0 40.6 L87.0 40.0 L88.0 39.3 L89.0 38.7 L90.0 38.1 L91.0 37.5 L92.0 36.9 L93.0 36.4 L94.0 35.9 L95.0 35.5 L96.0 35.0 L97.0 34.6 L98.0 34.3 L99.0 33.9 L100.0 33.6 L101.0 33.3 L102.0 33.0 L103.0 32.7 L104.0 32.5 L105.0 32.3 L106.0 32.1 L107.0 31.9 L108.0 31.7 L109.0 31.6 L110.0 31.4 L111.0 31.3 L112.0 31.2 L113.0 31.1 L114.0 31.0 L115.0 30.9 L116.0 30.8 L117.0 30.7 L118.0 30.7 L119.0 30.6 L120.0 30.5 L121.0 30.5 L122.0 30.4 L123.0 30.4 L124.0 30.4 L125.0 30.3 L126.0 30.3 L127.0 30.3 L128.0 30.2 L129.0 30.2 L130.0 30.2 L131.0 30.2 L132.0 30.2 L133.0 30.1 L134.0 30.1 L135.0 30.1 L136.0 30.1 L137.0 30.1 L138.0 30.1 L139.0 30.1 L140.0 30.1 L141.0 30.1 L142.0 30.1 L143.0 30.1 L144.0 30.0 L145.0 30.0 L146.0 30.0 L147.0 30.0 L148.0 30.0 L149.0 30.0 L150.0 30.0"/><path class="a-l" d="M10.0 78.0 L11.0 78.0 L12.0 78.0 L13.0 78.0 L14.0 78.0 L15.0 78.0 L16.0 78.0 L17.0 78.0 L18.0 78.0 L19.0 78.0 L20.0 78.0 L21.0 78.0 L22.0 78.0 L23.0 78.0 L24.0 78.0 L25.0 78.0 L26.0 78.0 L27.0 78.0 L28.0 78.0 L29.0 78.0 L30.0 78.0 L31.0 78.0 L32.0 78.0 L33.0 78.0 L34.0 77.9 L35.0 77.9 L36.0 77.9 L37.0 77.9 L38.0 77.9 L39.0 77.9 L40.0 77.9 L41.0 77.9 L42.0 77.8 L43.0 77.8 L44.0 77.8 L45.0 77.8 L46.0 77.7 L47.0 77.7 L48.0 77.6 L49.0 77.6 L50.0 77.5 L51.0 77.4 L52.0 77.4 L53.0 77.3 L54.0 77.1 L55.0 77.0 L56.0 76.9 L57.0 76.7 L58.0 76.5 L59.0 76.3 L60.0 76.0 L61.0 75.8 L62.0 75.4 L63.0 75.1 L64.0 74.7 L65.0 74.2 L66.0 73.7 L67.0 73.1 L68.0 72.5 L69.0 71.8 L70.0 71.0 L71.0 70.2 L72.0 69.3 L73.0 68.3 L74.0 67.3 L75.0 66.2 L76.0 65.0 L77.0 63.8 L78.0 62.6 L79.0 61.3 L80.0 60.0 L81.0 58.7 L82.0 57.4 L83.0 56.2 L84.0 55.0 L85.0 53.8 L86.0 52.7 L87.0 51.7 L88.0 50.7 L89.0 49.8 L90.0 49.0 L91.0 48.2 L92.0 47.5 L93.0 46.9 L94.0 46.3 L95.0 45.8 L96.0 45.3 L97.0 44.9 L98.0 44.6 L99.0 44.2 L100.0 44.0 L101.0 43.7 L102.0 43.5 L103.0 43.3 L104.0 43.1 L105.0 43.0 L106.0 42.9 L107.0 42.7 L108.0 42.6 L109.0 42.6 L110.0 42.5 L111.0 42.4 L112.0 42.4 L113.0 42.3 L114.0 42.3 L115.0 42.2 L116.0 42.2 L117.0 42.2 L118.0 42.2 L119.0 42.1 L120.0 42.1 L121.0 42.1 L122.0 42.1 L123.0 42.1 L124.0 42.1 L125.0 42.1 L126.0 42.1 L127.0 42.0 L128.0 42.0 L129.0 42.0 L130.0 42.0 L131.0 42.0 L132.0 42.0 L133.0 42.0 L134.0 42.0 L135.0 42.0 L136.0 42.0 L137.0 42.0 L138.0 42.0 L139.0 42.0 L140.0 42.0 L141.0 42.0 L142.0 42.0 L143.0 42.0 L144.0 42.0 L145.0 42.0 L146.0 42.0 L147.0 42.0 L148.0 42.0 L149.0 42.0 L150.0 42.0"/><path class="a-b" d="M10.0 60.0 L11.0 60.0 L12.0 60.0 L13.0 60.0 L14.0 60.0 L15.0 60.0 L16.0 60.0 L17.0 60.0 L18.0 60.0 L19.0 60.0 L20.0 60.0 L21.0 60.0 L22.0 60.0 L23.0 60.0 L24.0 60.0 L25.0 60.0 L26.0 60.0 L27.0 60.0 L28.0 60.0 L29.0 60.0 L30.0 60.0 L31.0 60.0 L32.0 60.0 L33.0 60.0 L34.0 60.0 L35.0 60.0 L36.0 60.0 L37.0 60.0 L38.0 60.0 L39.0 60.0 L40.0 60.0 L41.0 60.0 L42.0 60.0 L43.0 60.0 L44.0 60.0 L45.0 60.0 L46.0 60.0 L47.0 60.0 L48.0 60.0 L49.0 60.0 L50.0 60.0 L51.0 60.0 L52.0 60.0 L53.0 60.0 L54.0 60.0 L55.0 60.0 L56.0 60.0 L57.0 60.0 L58.0 60.0 L59.0 60.0 L60.0 60.0 L61.0 60.0 L62.0 60.0 L63.0 60.0 L64.0 60.0 L65.0 60.0 L66.0 60.0 L67.0 60.0 L68.0 60.0 L69.0 60.0 L70.0 60.0 L71.0 60.0 L72.0 60.0 L73.0 60.0 L74.0 60.0 L75.0 60.0 L76.0 60.0 L77.0 60.0 L78.0 60.0 L79.0 60.0 L80.0 60.0 L81.0 59.5 L82.0 58.9 L83.0 58.4 L84.0 57.8 L85.0 57.2 L86.0 56.7 L87.0 56.1 L88.0 55.6 L89.0 55.0 L90.0 54.5 L91.0 54.0 L92.0 53.4 L93.0 52.9 L94.0 52.3 L95.0 51.8 L96.0 51.2 L97.0 50.6 L98.0 50.1 L99.0 49.5 L100.0 49.0 L101.0 48.5 L102.0 47.9 L103.0 47.4 L104.0 46.8 L105.0 46.2 L106.0 45.7 L107.0 45.1 L108.0 44.6 L109.0 44.0 L110.0 43.5 L111.0 43.0 L112.0 42.4 L113.0 41.8 L114.0 41.3 L115.0 40.8 L116.0 40.2 L117.0 39.6 L118.0 39.1 L119.0 38.5 L120.0 38.0 L121.0 37.5 L122.0 36.9 L123.0 36.3 L124.0 35.8 L125.0 35.2 L126.0 34.7 L127.0 34.1 L128.0 33.6 L129.0 33.0 L130.0 32.5 L131.0 31.9 L132.0 31.4 L133.0 30.8 L134.0 30.3 L135.0 29.7 L136.0 29.2 L137.0 28.6 L138.0 28.1 L139.0 27.5 L140.0 27.0 L141.0 26.4 L142.0 25.9 L143.0 25.3 L144.0 24.8 L145.0 24.2 L146.0 23.7 L147.0 23.1 L148.0 22.6 L149.0 22.0 L150.0 21.5"/></svg></span><span class="al-meta"><span class="al-tag">Neural network basics</span><span>Beginner</span><span class="al-real">Real computation</span><span class="al-lab">Code lab</span></span><span class="al-t">Activation functions: ReLU, GELU, sigmoid, tanh and more</span><span class="al-d">Twelve activation functions side by side with their slopes: step, sigmoid, tanh, ReLU, Leaky ReLU, ELU, SELU, GELU, Swish, softplus and Mish.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/loss-functions" data-tag="neural network basics" data-q="loss functions: mse, cross-entropy, huber, focal and hinge loss functions"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><path class="a-l2" d="M10.0 78.0L150.0 78.0"/><path class="a-b" d="M10.0 12.0 L11.0 13.9 L12.0 15.7 L13.0 17.5 L14.0 19.3 L15.0 21.1 L16.0 22.8 L17.0 24.5 L18.0 26.2 L19.0 27.9 L20.0 29.5 L21.0 31.1 L22.0 32.7 L23.0 34.2 L24.0 35.8 L25.0 37.3 L26.0 38.7 L27.0 40.2 L28.0 41.6 L29.0 43.0 L30.0 44.3 L31.0 45.7 L32.0 47.0 L33.0 48.2 L34.0 49.5 L35.0 50.7 L36.0 51.9 L37.0 53.1 L38.0 54.2 L39.0 55.4 L40.0 56.4 L41.0 57.5 L42.0 58.6 L43.0 59.6 L44.0 60.5 L45.0 61.5 L46.0 62.4 L47.0 63.3 L48.0 64.2 L49.0 65.1 L50.0 65.9 L51.0 66.7 L52.0 67.4 L53.0 68.2 L54.0 68.9 L55.0 69.6 L56.0 70.2 L57.0 70.9 L58.0 71.5 L59.0 72.1 L60.0 72.6 L61.0 73.1 L62.0 73.6 L63.0 74.1 L64.0 74.6 L65.0 75.0 L66.0 75.4 L67.0 75.7 L68.0 76.1 L69.0 76.4 L70.0 76.7 L71.0 76.9 L72.0 77.1 L73.0 77.3 L74.0 77.5 L75.0 77.7 L76.0 77.8 L77.0 77.9 L78.0 77.9 L79.0 78.0 L80.0 78.0 L81.0 78.0 L82.0 77.9 L83.0 77.9 L84.0 77.8 L85.0 77.7 L86.0 77.5 L87.0 77.3 L88.0 77.1 L89.0 76.9 L90.0 76.7 L91.0 76.4 L92.0 76.1 L93.0 75.7 L94.0 75.4 L95.0 75.0 L96.0 74.6 L97.0 74.1 L98.0 73.6 L99.0 73.1 L100.0 72.6 L101.0 72.1 L102.0 71.5 L103.0 70.9 L104.0 70.2 L105.0 69.6 L106.0 68.9 L107.0 68.2 L108.0 67.4 L109.0 66.7 L110.0 65.9 L111.0 65.1 L112.0 64.2 L113.0 63.3 L114.0 62.4 L115.0 61.5 L116.0 60.5 L117.0 59.6 L118.0 58.6 L119.0 57.5 L120.0 56.4 L121.0 55.4 L122.0 54.2 L123.0 53.1 L124.0 51.9 L125.0 50.7 L126.0 49.5 L127.0 48.2 L128.0 47.0 L129.0 45.7 L130.0 44.3 L131.0 43.0 L132.0 41.6 L133.0 40.2 L134.0 38.7 L135.0 37.3 L136.0 35.8 L137.0 34.2 L138.0 32.7 L139.0 31.1 L140.0 29.5 L141.0 27.9 L142.0 26.2 L143.0 24.5 L144.0 22.8 L145.0 21.1 L146.0 19.3 L147.0 17.5 L148.0 15.7 L149.0 13.9 L150.0 12.0"/><path class="a-l" d="M10.0 38.0 L11.0 38.6 L12.0 39.1 L13.0 39.7 L14.0 40.3 L15.0 40.9 L16.0 41.4 L17.0 42.0 L18.0 42.6 L19.0 43.1 L20.0 43.7 L21.0 44.3 L22.0 44.9 L23.0 45.4 L24.0 46.0 L25.0 46.6 L26.0 47.1 L27.0 47.7 L28.0 48.3 L29.0 48.9 L30.0 49.4 L31.0 50.0 L32.0 50.6 L33.0 51.1 L34.0 51.7 L35.0 52.3 L36.0 52.9 L37.0 53.4 L38.0 54.0 L39.0 54.6 L40.0 55.1 L41.0 55.7 L42.0 56.3 L43.0 56.9 L44.0 57.4 L45.0 58.0 L46.0 58.6 L47.0 59.1 L48.0 59.7 L49.0 60.3 L50.0 60.9 L51.0 61.4 L52.0 62.0 L53.0 62.6 L54.0 63.1 L55.0 63.7 L56.0 64.3 L57.0 64.9 L58.0 65.4 L59.0 66.0 L60.0 66.6 L61.0 67.1 L62.0 67.7 L63.0 68.3 L64.0 68.9 L65.0 69.4 L66.0 70.0 L67.0 70.6 L68.0 71.1 L69.0 71.7 L70.0 72.3 L71.0 72.9 L72.0 73.4 L73.0 74.0 L74.0 74.6 L75.0 75.1 L76.0 75.7 L77.0 76.3 L78.0 76.9 L79.0 77.4 L80.0 78.0 L81.0 77.4 L82.0 76.9 L83.0 76.3 L84.0 75.7 L85.0 75.1 L86.0 74.6 L87.0 74.0 L88.0 73.4 L89.0 72.9 L90.0 72.3 L91.0 71.7 L92.0 71.1 L93.0 70.6 L94.0 70.0 L95.0 69.4 L96.0 68.9 L97.0 68.3 L98.0 67.7 L99.0 67.1 L100.0 66.6 L101.0 66.0 L102.0 65.4 L103.0 64.9 L104.0 64.3 L105.0 63.7 L106.0 63.1 L107.0 62.6 L108.0 62.0 L109.0 61.4 L110.0 60.9 L111.0 60.3 L112.0 59.7 L113.0 59.1 L114.0 58.6 L115.0 58.0 L116.0 57.4 L117.0 56.9 L118.0 56.3 L119.0 55.7 L120.0 55.1 L121.0 54.6 L122.0 54.0 L123.0 53.4 L124.0 52.9 L125.0 52.3 L126.0 51.7 L127.0 51.1 L128.0 50.6 L129.0 50.0 L130.0 49.4 L131.0 48.9 L132.0 48.3 L133.0 47.7 L134.0 47.1 L135.0 46.6 L136.0 46.0 L137.0 45.4 L138.0 44.9 L139.0 44.3 L140.0 43.7 L141.0 43.1 L142.0 42.6 L143.0 42.0 L144.0 41.4 L145.0 40.9 L146.0 40.3 L147.0 39.7 L148.0 39.1 L149.0 38.6 L150.0 38.0"/><path class="a-l2" d="M10.0 27.5 L11.0 28.4 L12.0 29.3 L13.0 30.2 L14.0 31.1 L15.0 31.9 L16.0 32.8 L17.0 33.7 L18.0 34.6 L19.0 35.5 L20.0 36.4 L21.0 37.3 L22.0 38.2 L23.0 39.1 L24.0 40.0 L25.0 40.9 L26.0 41.8 L27.0 42.7 L28.0 43.6 L29.0 44.5 L30.0 45.4 L31.0 46.2 L32.0 47.1 L33.0 48.0 L34.0 48.9 L35.0 49.8 L36.0 50.7 L37.0 51.5 L38.0 52.4 L39.0 53.3 L40.0 54.1 L41.0 55.0 L42.0 55.9 L43.0 56.7 L44.0 57.6 L45.0 58.4 L46.0 59.3 L47.0 60.1 L48.0 61.0 L49.0 61.8 L50.0 62.6 L51.0 63.4 L52.0 64.2 L53.0 65.0 L54.0 65.8 L55.0 66.6 L56.0 67.3 L57.0 68.1 L58.0 68.8 L59.0 69.5 L60.0 70.2 L61.0 70.9 L62.0 71.5 L63.0 72.2 L64.0 72.8 L65.0 73.4 L66.0 73.9 L67.0 74.4 L68.0 74.9 L69.0 75.4 L70.0 75.8 L71.0 76.2 L72.0 76.6 L73.0 76.9 L74.0 77.2 L75.0 77.4 L76.0 77.6 L77.0 77.8 L78.0 77.9 L79.0 78.0 L80.0 78.0 L81.0 78.0 L82.0 77.9 L83.0 77.8 L84.0 77.6 L85.0 77.4 L86.0 77.2 L87.0 76.9 L88.0 76.6 L89.0 76.2 L90.0 75.8 L91.0 75.4 L92.0 74.9 L93.0 74.4 L94.0 73.9 L95.0 73.4 L96.0 72.8 L97.0 72.2 L98.0 71.5 L99.0 70.9 L100.0 70.2 L101.0 69.5 L102.0 68.8 L103.0 68.1 L104.0 67.3 L105.0 66.6 L106.0 65.8 L107.0 65.0 L108.0 64.2 L109.0 63.4 L110.0 62.6 L111.0 61.8 L112.0 61.0 L113.0 60.1 L114.0 59.3 L115.0 58.4 L116.0 57.6 L117.0 56.7 L118.0 55.9 L119.0 55.0 L120.0 54.1 L121.0 53.3 L122.0 52.4 L123.0 51.5 L124.0 50.7 L125.0 49.8 L126.0 48.9 L127.0 48.0 L128.0 47.1 L129.0 46.2 L130.0 45.4 L131.0 44.5 L132.0 43.6 L133.0 42.7 L134.0 41.8 L135.0 40.9 L136.0 40.0 L137.0 39.1 L138.0 38.2 L139.0 37.3 L140.0 36.4 L141.0 35.5 L142.0 34.6 L143.0 33.7 L144.0 32.8 L145.0 31.9 L146.0 31.1 L147.0 30.2 L148.0 29.3 L149.0 28.4 L150.0 27.5"/></svg></span><span class="al-meta"><span class="al-tag">Neural network basics</span><span>Beginner</span><span class="al-real">Real computation</span><span class="al-lab">Code lab</span></span><span class="al-t">Loss functions: MSE, cross-entropy, Huber, focal and hinge</span><span class="al-d">Eight loss functions and their gradients: MSE, MAE, Huber, log-cosh, binary and categorical cross-entropy, focal loss and hinge, all interactive.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/backpropagation" data-tag="neural network basics" data-q="backpropagation, one number at a time backpropagation"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><path class="a-l" d="M20.0 38.0L80.0 38.0"/><path class="a-l" d="M20.0 38.0L80.0 52.0"/><path class="a-l" d="M20.0 52.0L80.0 38.0"/><path class="a-b" d="M20.0 52.0L80.0 52.0"/><path class="a-l" d="M80.0 38.0L140.0 45.0"/><path class="a-l" d="M80.0 52.0L140.0 45.0"/><circle class="a-db" cx="20.0" cy="38.0" r="1.8"/><circle class="a-d" cx="20.0" cy="52.0" r="1.8"/><circle class="a-d" cx="80.0" cy="38.0" r="1.8"/><circle class="a-db" cx="80.0" cy="52.0" r="1.8"/><circle class="a-db" cx="140.0" cy="45.0" r="1.8"/><path class="a-b" d="M140.0 45.0L80.0 38.0"/><path class="a-b" d="M140.0 45.0L80.0 52.0"/></svg></span><span class="al-meta"><span class="al-tag">Neural network basics</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-lab">Code lab</span></span><span class="al-t">Backpropagation, one number at a time</span><span class="al-d">Step through the forward pass, loss, backward pass and weight update of a tiny network, with every value and gradient shown on the graph.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/gradient-descent-optimisers" data-tag="neural network basics" data-q="gradient descent and its variants: sgd, momentum, nesterov, rmsprop, adam optimisers"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><ellipse class="a-l2" cx="92" cy="46" rx="8" ry="3"/><ellipse class="a-l2" cx="92" cy="46" rx="16" ry="7"/><ellipse class="a-l2" cx="92" cy="46" rx="24" ry="10"/><ellipse class="a-l2" cx="92" cy="46" rx="32" ry="14"/><ellipse class="a-l2" cx="92" cy="46" rx="40" ry="18"/><ellipse class="a-l2" cx="92" cy="46" rx="48" ry="21"/><ellipse class="a-l2" cx="92" cy="46" rx="56" ry="25"/><ellipse class="a-l2" cx="92" cy="46" rx="64" ry="28"/><path class="a-l" d="M22.0 26.0 L34.0 62.0 L44.0 30.0 L56.0 58.0 L66.0 36.0 L76.0 52.0 L86.0 42.0 L92.0 46.0"/><path class="a-b" d="M22.0 26.0 L50.0 40.0 L72.0 45.0 L92.0 46.0"/><circle class="a-db" cx="92.0" cy="46.0" r="2.0"/></svg></span><span class="al-meta"><span class="al-tag">Neural network basics</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-lab">Code lab</span></span><span class="al-t">Gradient descent and its variants: SGD, momentum, Nesterov, RMSProp, Adam</span><span class="al-d">Six optimisers race on the same loss surface: SGD, momentum, Nesterov, AdaGrad, RMSProp and Adam, with paths, contours and loss per step.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/convolution" data-tag="neural network basics" data-q="convolution: how a cnn sees edges convolution"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-g" x="14.0" y="12.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="14.0" y="18.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="14.0" y="25.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="14.0" y="31.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="14.0" y="38.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="14.0" y="44.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="14.0" y="51.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="14.0" y="57.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="14.0" y="64.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="14.0" y="70.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="20.5" y="12.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="20.5" y="18.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="20.5" y="25.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="20.5" y="31.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="20.5" y="38.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="20.5" y="44.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="20.5" y="51.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="20.5" y="57.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="20.5" y="64.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="20.5" y="70.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="27.0" y="12.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="27.0" y="18.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="27.0" y="25.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="27.0" y="31.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="27.0" y="38.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="27.0" y="44.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="27.0" y="51.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="27.0" y="57.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="27.0" y="64.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="27.0" y="70.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="33.5" y="12.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="33.5" y="18.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="33.5" y="25.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="33.5" y="31.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="33.5" y="38.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="33.5" y="44.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="33.5" y="51.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="33.5" y="57.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="33.5" y="64.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="33.5" y="70.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="40.0" y="12.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="40.0" y="18.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="40.0" y="25.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="40.0" y="31.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="40.0" y="38.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="40.0" y="44.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="40.0" y="51.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="40.0" y="57.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="40.0" y="64.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="40.0" y="70.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="46.5" y="12.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="46.5" y="18.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="46.5" y="25.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="46.5" y="31.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="46.5" y="38.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="46.5" y="44.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="46.5" y="51.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="46.5" y="57.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="46.5" y="64.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="46.5" y="70.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="53.0" y="12.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="53.0" y="18.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="53.0" y="25.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="53.0" y="31.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="53.0" y="38.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="53.0" y="44.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="53.0" y="51.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="53.0" y="57.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="53.0" y="64.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="53.0" y="70.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="59.5" y="12.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="59.5" y="18.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="59.5" y="25.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="59.5" y="31.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="59.5" y="38.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="59.5" y="44.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="59.5" y="51.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="59.5" y="57.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="59.5" y="64.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="59.5" y="70.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="66.0" y="12.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="66.0" y="18.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="66.0" y="25.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="66.0" y="31.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="66.0" y="38.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="66.0" y="44.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="66.0" y="51.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="66.0" y="57.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="66.0" y="64.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="66.0" y="70.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="72.5" y="12.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="72.5" y="18.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="72.5" y="25.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="72.5" y="31.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="72.5" y="38.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="72.5" y="44.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="72.5" y="51.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="72.5" y="57.5" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="72.5" y="64.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="72.5" y="70.5" width="5.5" height="5.5" rx="1"/><rect class="a-f" x="33.5" y="31.5" width="18.5" height="18.5" rx="1"/><path class="a-b" d="M53.0 41.0L100.0 41.0"/><rect class="a-g" x="104.0" y="22.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="104.0" y="27.5" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="104.0" y="33.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="104.0" y="38.5" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="104.0" y="44.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="104.0" y="49.5" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="104.0" y="55.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="109.5" y="22.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="109.5" y="27.5" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="109.5" y="33.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="109.5" y="38.5" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="109.5" y="44.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="109.5" y="49.5" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="109.5" y="55.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="115.0" y="22.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="115.0" y="27.5" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="115.0" y="33.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="115.0" y="38.5" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="115.0" y="44.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="115.0" y="49.5" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="115.0" y="55.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="120.5" y="22.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="120.5" y="27.5" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="120.5" y="33.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="120.5" y="38.5" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="120.5" y="44.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="120.5" y="49.5" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="120.5" y="55.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="126.0" y="22.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="126.0" y="27.5" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="126.0" y="33.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="126.0" y="38.5" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="126.0" y="44.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="126.0" y="49.5" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="126.0" y="55.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="131.5" y="22.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="131.5" y="27.5" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="131.5" y="33.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="131.5" y="38.5" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="131.5" y="44.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="131.5" y="49.5" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="131.5" y="55.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="137.0" y="22.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="137.0" y="27.5" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="137.0" y="33.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="137.0" y="38.5" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="137.0" y="44.0" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="137.0" y="49.5" width="4.6" height="4.6" rx="1"/><rect class="a-g" x="137.0" y="55.0" width="4.6" height="4.6" rx="1"/><rect class="a-f" x="108.0" y="37.0" width="4.6" height="4.6" rx="1"/></svg></span><span class="al-meta"><span class="al-tag">Neural network basics</span><span>Beginner</span><span class="al-real">Real computation</span><span class="al-lab">Code lab</span></span><span class="al-t">Convolution: how a CNN sees edges</span><span class="al-d">Slide a 3×3 kernel over a handwritten digit and build the feature map pixel by pixel: edge, blur and sharpen filters, editable weights.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a></div></section><section class="al-month"><h2 class="al-mh" id="training">Training <span>4</span></h2><div class="al-grid"><a class="al-card" href="/ai-simulations/neural-network-playground" data-tag="training" data-q="neural network playground: watch every neuron learn network playground"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><path class="a-l2" d="M20.0 38.0L60.0 21.0"/><path class="a-l2" d="M20.0 38.0L60.0 33.0"/><path class="a-l" d="M20.0 38.0L60.0 45.0"/><path class="a-b" d="M20.0 38.0L60.0 57.0"/><path class="a-l" d="M20.0 38.0L60.0 69.0"/><path class="a-l2" d="M20.0 52.0L60.0 21.0"/><path class="a-l2" d="M20.0 52.0L60.0 33.0"/><path class="a-l2" d="M20.0 52.0L60.0 45.0"/><path class="a-l2" d="M20.0 52.0L60.0 57.0"/><path class="a-l2" d="M20.0 52.0L60.0 69.0"/><path class="a-l2" d="M60.0 21.0L100.0 21.0"/><path class="a-b" d="M60.0 21.0L100.0 33.0"/><path class="a-l2" d="M60.0 21.0L100.0 45.0"/><path class="a-l" d="M60.0 21.0L100.0 57.0"/><path class="a-l" d="M60.0 21.0L100.0 69.0"/><path class="a-l2" d="M60.0 33.0L100.0 21.0"/><path class="a-l2" d="M60.0 33.0L100.0 33.0"/><path class="a-l2" d="M60.0 33.0L100.0 45.0"/><path class="a-l2" d="M60.0 33.0L100.0 57.0"/><path class="a-l" d="M60.0 33.0L100.0 69.0"/><path class="a-l" d="M60.0 45.0L100.0 21.0"/><path class="a-l" d="M60.0 45.0L100.0 33.0"/><path class="a-l" d="M60.0 45.0L100.0 45.0"/><path class="a-l" d="M60.0 45.0L100.0 57.0"/><path class="a-l2" d="M60.0 45.0L100.0 69.0"/><path class="a-l2" d="M60.0 57.0L100.0 21.0"/><path class="a-l2" d="M60.0 57.0L100.0 33.0"/><path class="a-l2" d="M60.0 57.0L100.0 45.0"/><path class="a-l" d="M60.0 57.0L100.0 57.0"/><path class="a-l2" d="M60.0 57.0L100.0 69.0"/><path class="a-l2" d="M60.0 69.0L100.0 21.0"/><path class="a-l" d="M60.0 69.0L100.0 33.0"/><path class="a-l2" d="M60.0 69.0L100.0 45.0"/><path class="a-l" d="M60.0 69.0L100.0 57.0"/><path class="a-l2" d="M60.0 69.0L100.0 69.0"/><path class="a-b" d="M100.0 21.0L140.0 45.0"/><path class="a-l" d="M100.0 33.0L140.0 45.0"/><path class="a-l" d="M100.0 45.0L140.0 45.0"/><path class="a-l2" d="M100.0 57.0L140.0 45.0"/><path class="a-b" d="M100.0 69.0L140.0 45.0"/><circle class="a-d" cx="20.0" cy="38.0" r="1.8"/><circle class="a-db" cx="20.0" cy="52.0" r="1.8"/><circle class="a-d" cx="60.0" cy="21.0" r="1.8"/><circle class="a-d" cx="60.0" cy="33.0" r="1.8"/><circle class="a-d" cx="60.0" cy="45.0" r="1.8"/><circle class="a-d" cx="60.0" cy="57.0" r="1.8"/><circle class="a-d" cx="60.0" cy="69.0" r="1.8"/><circle class="a-d" cx="100.0" cy="21.0" r="1.8"/><circle class="a-d" cx="100.0" cy="33.0" r="1.8"/><circle class="a-db" cx="100.0" cy="45.0" r="1.8"/><circle class="a-d" cx="100.0" cy="57.0" r="1.8"/><circle class="a-db" cx="100.0" cy="69.0" r="1.8"/><circle class="a-db" cx="140.0" cy="45.0" r="1.8"/></svg></span><span class="al-meta"><span class="al-tag">Training</span><span>Beginner</span><span class="al-real">Real computation</span><span class="al-lab">Code lab</span></span><span class="al-t">Neural network playground: watch every neuron learn</span><span class="al-d">Build a small neural network, train it on four 2D datasets and see what every neuron learns, with live weights, loss and decision boundary.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/overfitting-and-regularisation" data-tag="training" data-q="overfitting and regularisation: when a model learns the noise overfitting"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><path class="a-l2" d="M10.0 40.3 L11.0 39.5 L12.0 38.6 L13.0 37.7 L14.0 36.9 L15.0 36.1 L16.0 35.4 L17.0 34.6 L18.0 33.9 L19.0 33.3 L20.0 32.6 L21.0 32.0 L22.0 31.5 L23.0 31.0 L24.0 30.5 L25.0 30.0 L26.0 29.6 L27.0 29.3 L28.0 29.0 L29.0 28.7 L30.0 28.5 L31.0 28.3 L32.0 28.1 L33.0 28.1 L34.0 28.0 L35.0 28.0 L36.0 28.0 L37.0 28.1 L38.0 28.3 L39.0 28.4 L40.0 28.7 L41.0 28.9 L42.0 29.2 L43.0 29.6 L44.0 30.0 L45.0 30.4 L46.0 30.9 L47.0 31.4 L48.0 32.0 L49.0 32.6 L50.0 33.2 L51.0 33.9 L52.0 34.6 L53.0 35.3 L54.0 36.0 L55.0 36.8 L56.0 37.6 L57.0 38.5 L58.0 39.4 L59.0 40.2 L60.0 41.1 L61.0 42.1 L62.0 43.0 L63.0 44.0 L64.0 44.9 L65.0 45.9 L66.0 46.9 L67.0 47.9 L68.0 48.9 L69.0 49.9 L70.0 50.9 L71.0 51.9 L72.0 52.9 L73.0 53.9 L74.0 54.8 L75.0 55.8 L76.0 56.8 L77.0 57.7 L78.0 58.6 L79.0 59.6 L80.0 60.4 L81.0 61.3 L82.0 62.2 L83.0 63.0 L84.0 63.8 L85.0 64.5 L86.0 65.3 L87.0 66.0 L88.0 66.6 L89.0 67.3 L90.0 67.9 L91.0 68.4 L92.0 69.0 L93.0 69.5 L94.0 69.9 L95.0 70.3 L96.0 70.7 L97.0 71.0 L98.0 71.3 L99.0 71.5 L100.0 71.7 L101.0 71.8 L102.0 71.9 L103.0 72.0 L104.0 72.0 L105.0 72.0 L106.0 71.9 L107.0 71.7 L108.0 71.6 L109.0 71.4 L110.0 71.1 L111.0 70.8 L112.0 70.4 L113.0 70.1 L114.0 69.6 L115.0 69.1 L116.0 68.6 L117.0 68.1 L118.0 67.5 L119.0 66.9 L120.0 66.2 L121.0 65.5 L122.0 64.8 L123.0 64.0 L124.0 63.3 L125.0 62.4 L126.0 61.6 L127.0 60.7 L128.0 59.9 L129.0 59.0 L130.0 58.0 L131.0 57.1 L132.0 56.1 L133.0 55.2 L134.0 54.2 L135.0 53.2 L136.0 52.2 L137.0 51.2 L138.0 50.2 L139.0 49.2 L140.0 48.2 L141.0 47.2 L142.0 46.2 L143.0 45.3 L144.0 44.3 L145.0 43.3 L146.0 42.4 L147.0 41.5 L148.0 40.5 L149.0 39.7"/><path class="a-b" d="M10.0 40.3 L11.0 39.5 L12.0 38.7 L13.0 37.9 L14.0 37.2 L15.0 36.5 L16.0 35.7 L17.0 35.0 L18.0 34.2 L19.0 33.5 L20.0 32.7 L21.0 31.8 L22.0 31.0 L23.0 30.3 L24.0 29.6 L25.0 29.1 L26.0 28.7 L27.0 28.4 L28.0 28.3 L29.0 28.3 L30.0 28.4 L31.0 28.7 L32.0 28.9 L33.0 29.2 L34.0 29.5 L35.0 29.6 L36.0 29.7 L37.0 29.6 L38.0 29.4 L39.0 29.1 L40.0 28.8 L41.0 28.4 L42.0 28.1 L43.0 28.0 L44.0 28.0 L45.0 28.2 L46.0 28.7 L47.0 29.4 L48.0 30.4 L49.0 31.6 L50.0 33.0 L51.0 34.5 L52.0 36.0 L53.0 37.4 L54.0 38.7 L55.0 39.7 L56.0 40.5 L57.0 41.1 L58.0 41.4 L59.0 41.5 L60.0 41.4 L61.0 41.3 L62.0 41.3 L63.0 41.4 L64.0 41.7 L65.0 42.4 L66.0 43.4 L67.0 44.7 L68.0 46.4 L69.0 48.4 L70.0 50.5 L71.0 52.7 L72.0 54.9 L73.0 56.9 L74.0 58.6 L75.0 60.0 L76.0 60.9 L77.0 61.5 L78.0 61.6 L79.0 61.4 L80.0 61.0 L81.0 60.4 L82.0 59.9 L83.0 59.5 L84.0 59.5 L85.0 59.8 L86.0 60.5 L87.0 61.6 L88.0 63.2 L89.0 65.1 L90.0 67.2 L91.0 69.4 L92.0 71.5 L93.0 73.3 L94.0 74.8 L95.0 75.7 L96.0 76.1 L97.0 75.9 L98.0 75.2 L99.0 74.1 L100.0 72.6 L101.0 70.9 L102.0 69.2 L103.0 67.8 L104.0 66.6 L105.0 65.9 L106.0 65.8 L107.0 66.2 L108.0 67.1 L109.0 68.4 L110.0 70.0 L111.0 71.7 L112.0 73.3 L113.0 74.7 L114.0 75.5 L115.0 75.8 L116.0 75.4 L117.0 74.3 L118.0 72.5 L119.0 70.2 L120.0 67.5 L121.0 64.6 L122.0 61.7 L123.0 59.1 L124.0 56.8 L125.0 55.2 L126.0 54.2 L127.0 53.9 L128.0 54.3 L129.0 55.2 L130.0 56.5 L131.0 58.0 L132.0 59.4 L133.0 60.5 L134.0 61.1 L135.0 61.1 L136.0 60.3 L137.0 58.7 L138.0 56.4 L139.0 53.4 L140.0 50.0 L141.0 46.4 L142.0 42.9 L143.0 39.6 L144.0 36.9 L145.0 34.9 L146.0 33.7 L147.0 33.4 L148.0 33.8 L149.0 35.0"/><circle class="a-d" cx="14.0" cy="40.7" r="1.4"/><circle class="a-d" cx="23.5" cy="35.9" r="1.4"/><circle class="a-d" cx="33.0" cy="37.4" r="1.4"/><circle class="a-d" cx="42.5" cy="29.8" r="1.4"/><circle class="a-d" cx="52.0" cy="34.7" r="1.4"/><circle class="a-d" cx="61.5" cy="37.8" r="1.4"/><circle class="a-d" cx="71.0" cy="53.9" r="1.4"/><circle class="a-d" cx="80.5" cy="59.9" r="1.4"/><circle class="a-d" cx="90.0" cy="69.4" r="1.4"/><circle class="a-d" cx="99.5" cy="67.8" r="1.4"/><circle class="a-d" cx="109.0" cy="67.5" r="1.4"/><circle class="a-d" cx="118.5" cy="62.3" r="1.4"/><circle class="a-d" cx="128.0" cy="54.1" r="1.4"/><circle class="a-d" cx="137.5" cy="50.4" r="1.4"/></svg></span><span class="al-meta"><span class="al-tag">Training</span><span>Beginner</span><span class="al-real">Real computation</span><span class="al-lab">Code lab</span></span><span class="al-t">Overfitting and regularisation: when a model learns the noise</span><span class="al-d">Fit polynomials of growing degree to noisy points and watch training error fall while test error rises, then fix it with more data or an L2 penalty.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/normalisation-and-dropout" data-tag="training" data-q="initialisation, batch normalisation and dropout in a deep network init, batchnorm, dropout"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><path class="a-l2" d="M10.0 20.0L150.0 20.0"/><path class="a-l2" d="M10.0 38.0L150.0 38.0"/><path class="a-l2" d="M10.0 56.0L150.0 56.0"/><path class="a-l2" d="M10.0 74.0L150.0 74.0"/><path class="a-l" d="M14.0 46.0 L20.6 44.4 L27.2 42.8 L33.8 41.2 L40.4 39.6 L47.0 38.0 L53.6 36.4 L60.2 34.8 L66.8 33.2 L73.4 31.6 L80.0 30.0 L86.6 28.4 L93.2 26.8 L99.8 25.2 L106.4 23.6 L113.0 22.0 L119.6 20.4 L126.2 18.8 L132.8 17.2 L139.4 15.6"/><path class="a-l2" d="M14.0 46.0 L20.6 47.5 L27.2 49.0 L33.8 50.5 L40.4 52.0 L47.0 53.5 L53.6 55.0 L60.2 56.5 L66.8 58.0 L73.4 59.5 L80.0 61.0 L86.6 62.5 L93.2 64.0 L99.8 65.5 L106.4 67.0 L113.0 68.5 L119.6 70.0 L126.2 71.5 L132.8 73.0 L139.4 74.5"/><path class="a-b" d="M14.0 46.0 L20.6 47.0 L27.2 47.1 L33.8 46.2 L40.4 45.1 L47.0 44.8 L53.6 45.7 L60.2 46.8 L66.8 47.2 L73.4 46.5 L80.0 45.3 L86.6 44.8 L93.2 45.4 L99.8 46.5 L106.4 47.2 L113.0 46.8 L119.6 45.7 L126.2 44.8 L132.8 45.1 L139.4 46.2"/><circle class="a-db" cx="14.0" cy="46.0" r="1.2"/><circle class="a-db" cx="33.8" cy="46.2" r="1.2"/><circle class="a-db" cx="53.6" cy="45.7" r="1.2"/><circle class="a-db" cx="73.4" cy="46.5" r="1.2"/><circle class="a-db" cx="93.2" cy="45.4" r="1.2"/><circle class="a-db" cx="113.0" cy="46.8" r="1.2"/><circle class="a-db" cx="132.8" cy="45.1" r="1.2"/></svg></span><span class="al-meta"><span class="al-tag">Training</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-lab">Code lab</span></span><span class="al-t">Initialisation, batch normalisation and dropout in a deep network</span><span class="al-d">Push a real batch through up to 40 layers and watch activations vanish or explode with bad initialisation, and stay healthy with He init or batch norm.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/learning-rate-schedules" data-tag="training" data-q="learning-rate schedules: warmup, cosine, step and one-cycle lr schedules"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><path class="a-l2" d="M10.0 80.0L150.0 80.0"/><path class="a-b" d="M12.0 80.0 L13.4 74.0 L14.7 68.0 L16.1 62.0 L17.4 56.0 L18.8 50.0 L20.2 44.0 L21.5 38.0 L22.9 32.0 L24.2 26.0 L25.6 20.0 L27.0 20.0 L28.3 20.1 L29.7 20.2 L31.0 20.3 L32.4 20.5 L33.8 20.7 L35.1 20.9 L36.5 21.2 L37.8 21.5 L39.2 21.8 L40.6 22.2 L41.9 22.6 L43.3 23.0 L44.6 23.5 L46.0 24.0 L47.4 24.6 L48.7 25.1 L50.1 25.7 L51.4 26.4 L52.8 27.0 L54.2 27.7 L55.5 28.4 L56.9 29.2 L58.2 29.9 L59.6 30.7 L61.0 31.5 L62.3 32.4 L63.7 33.2 L65.0 34.1 L66.4 35.0 L67.8 35.9 L69.1 36.8 L70.5 37.8 L71.8 38.8 L73.2 39.7 L74.6 40.7 L75.9 41.7 L77.3 42.7 L78.6 43.8 L80.0 44.8 L81.4 45.8 L82.7 46.9 L84.1 47.9 L85.4 49.0 L86.8 50.0 L88.2 51.0 L89.5 52.1 L90.9 53.1 L92.2 54.2 L93.6 55.2 L95.0 56.2 L96.3 57.3 L97.7 58.3 L99.0 59.3 L100.4 60.3 L101.8 61.2 L103.1 62.2 L104.5 63.2 L105.8 64.1 L107.2 65.0 L108.6 65.9 L109.9 66.8 L111.3 67.6 L112.6 68.5 L114.0 69.3 L115.4 70.1 L116.7 70.8 L118.1 71.6 L119.4 72.3 L120.8 73.0 L122.2 73.6 L123.5 74.3 L124.9 74.9 L126.2 75.4 L127.6 76.0 L129.0 76.5 L130.3 77.0 L131.7 77.4 L133.0 77.8 L134.4 78.2 L135.8 78.5 L137.1 78.8 L138.5 79.1 L139.8 79.3 L141.2 79.5 L142.6 79.7 L143.9 79.8 L145.3 79.9 L146.6 80.0 L148.0 80.0"/><path class="a-l" d="M12.0 20.0 L13.4 20.0 L14.7 20.0 L16.1 20.0 L17.4 20.0 L18.8 20.0 L20.2 20.0 L21.5 20.0 L22.9 20.0 L24.2 20.0 L25.6 20.0 L27.0 20.0 L28.3 20.0 L29.7 20.0 L31.0 20.0 L32.4 20.0 L33.8 20.0 L35.1 20.0 L36.5 20.0 L37.8 20.0 L39.2 20.0 L40.6 20.0 L41.9 20.0 L43.3 20.0 L44.6 20.0 L46.0 20.0 L47.4 20.0 L48.7 20.0 L50.1 20.0 L51.4 20.0 L52.8 20.0 L54.2 20.0 L55.5 20.0 L56.9 20.0 L58.2 62.0 L59.6 62.0 L61.0 62.0 L62.3 62.0 L63.7 62.0 L65.0 62.0 L66.4 62.0 L67.8 62.0 L69.1 62.0 L70.5 62.0 L71.8 62.0 L73.2 62.0 L74.6 62.0 L75.9 62.0 L77.3 62.0 L78.6 62.0 L80.0 62.0 L81.4 62.0 L82.7 62.0 L84.1 62.0 L85.4 62.0 L86.8 62.0 L88.2 62.0 L89.5 62.0 L90.9 62.0 L92.2 62.0 L93.6 62.0 L95.0 62.0 L96.3 62.0 L97.7 62.0 L99.0 62.0 L100.4 62.0 L101.8 62.0 L103.1 62.0 L104.5 74.6 L105.8 74.6 L107.2 74.6 L108.6 74.6 L109.9 74.6 L111.3 74.6 L112.6 74.6 L114.0 74.6 L115.4 74.6 L116.7 74.6 L118.1 74.6 L119.4 74.6 L120.8 74.6 L122.2 74.6 L123.5 74.6 L124.9 74.6 L126.2 74.6 L127.6 74.6 L129.0 74.6 L130.3 74.6 L131.7 74.6 L133.0 74.6 L134.4 74.6 L135.8 74.6 L137.1 74.6 L138.5 74.6 L139.8 74.6 L141.2 74.6 L142.6 74.6 L143.9 74.6 L145.3 74.6 L146.6 74.6 L148.0 74.6"/><path class="a-l2" d="M12.0 80.0 L13.4 78.0 L14.7 76.0 L16.1 74.0 L17.4 72.0 L18.8 70.0 L20.2 68.0 L21.5 66.0 L22.9 64.0 L24.2 62.0 L25.6 60.0 L27.0 58.0 L28.3 56.0 L29.7 54.0 L31.0 52.0 L32.4 50.0 L33.8 48.0 L35.1 46.0 L36.5 44.0 L37.8 42.0 L39.2 40.0 L40.6 38.0 L41.9 36.0 L43.3 34.0 L44.6 32.0 L46.0 30.0 L47.4 28.0 L48.7 26.0 L50.1 24.0 L51.4 22.0 L52.8 20.0 L54.2 20.9 L55.5 21.7 L56.9 22.6 L58.2 23.4 L59.6 24.3 L61.0 25.1 L62.3 26.0 L63.7 26.9 L65.0 27.7 L66.4 28.6 L67.8 29.4 L69.1 30.3 L70.5 31.1 L71.8 32.0 L73.2 32.9 L74.6 33.7 L75.9 34.6 L77.3 35.4 L78.6 36.3 L80.0 37.1 L81.4 38.0 L82.7 38.9 L84.1 39.7 L85.4 40.6 L86.8 41.4 L88.2 42.3 L89.5 43.1 L90.9 44.0 L92.2 44.9 L93.6 45.7 L95.0 46.6 L96.3 47.4 L97.7 48.3 L99.0 49.1 L100.4 50.0 L101.8 50.9 L103.1 51.7 L104.5 52.6 L105.8 53.4 L107.2 54.3 L108.6 55.1 L109.9 56.0 L111.3 56.9 L112.6 57.7 L114.0 58.6 L115.4 59.4 L116.7 60.3 L118.1 61.1 L119.4 62.0 L120.8 62.9 L122.2 63.7 L123.5 64.6 L124.9 65.4 L126.2 66.3 L127.6 67.1 L129.0 68.0 L130.3 68.9 L131.7 69.7 L133.0 70.6 L134.4 71.4 L135.8 72.3 L137.1 73.1 L138.5 74.0 L139.8 74.9 L141.2 75.7 L142.6 76.6 L143.9 77.4 L145.3 78.3 L146.6 79.1 L148.0 80.0"/></svg></span><span class="al-meta"><span class="al-tag">Training</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-lab">Code lab</span></span><span class="al-t">Learning-rate schedules: warmup, cosine, step and one-cycle</span><span class="al-d">Compare eight learning-rate schedules and see their effect on a real noisy optimisation: warmup, cosine, step decay, one-cycle, linear and inverse square root.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a></div></section><section class="al-month"><h2 class="al-mh" id="language-models">Language models <span>9</span></h2><div class="al-grid"><a class="al-card" href="/ai-simulations/tokenisation-bpe" data-tag="language models" data-q="tokenisation with byte-pair encoding, merge by merge tokenisation"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-g" x="10.0" y="20.0" width="16.0" height="11.0" rx="2"/><text class="a-t" x="12.0" y="27.5">the</text><rect class="a-g" x="28.0" y="20.0" width="6.0" height="11.0" rx="2"/><text class="a-t" x="30.0" y="27.5">␣</text><rect class="a-g" x="38.0" y="20.0" width="22.0" height="11.0" rx="2"/><text class="a-t" x="40.0" y="27.5">low</text><rect class="a-g" x="62.0" y="20.0" width="12.0" height="11.0" rx="2"/><text class="a-t" x="64.0" y="27.5">er</text><rect class="a-g" x="76.0" y="20.0" width="6.0" height="11.0" rx="2"/><text class="a-t" x="78.0" y="27.5">␣</text><rect class="a-g" x="86.0" y="20.0" width="24.0" height="11.0" rx="2"/><text class="a-t" x="88.0" y="27.5">new</text><rect class="a-g" x="112.0" y="20.0" width="14.0" height="11.0" rx="2"/><text class="a-t" x="114.0" y="27.5">est</text><rect class="a-g" x="10.0" y="52.0" width="28.0" height="11.0" rx="2"/><text class="a-t" x="12.0" y="59.5">learn</text><rect class="a-g" x="40.0" y="52.0" width="14.0" height="11.0" rx="2"/><text class="a-t" x="42.0" y="59.5">ing</text><rect class="a-g" x="58.0" y="52.0" width="30.0" height="11.0" rx="2"/><text class="a-t" x="60.0" y="59.5">slow</text><rect class="a-g" x="90.0" y="52.0" width="14.0" height="11.0" rx="2"/><text class="a-t" x="92.0" y="59.5">est</text><rect class="a-f" x="102.0" y="20.0" width="14.0" height="11.0" rx="2"/><rect class="a-f" x="80.0" y="52.0" width="14.0" height="11.0" rx="2"/></svg></span><span class="al-meta"><span class="al-tag">Language models</span><span>Beginner</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Tokenisation with byte-pair encoding, merge by merge</span><span class="al-d">See how a tokeniser is trained: byte-pair encoding counts adjacent pairs and merges the most frequent one, building a vocabulary step by step.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/attention" data-tag="language models" data-q="attention: how a transformer decides what to look at attention"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-f" x="40.0" y="8.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="40.0" y="16.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="49.0" y="16.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="40.0" y="24.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="49.0" y="24.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="58.0" y="24.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="40.0" y="32.0" width="8.0" height="7.0" rx="1"/><rect class="a-f" x="49.0" y="32.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="58.0" y="32.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="67.0" y="32.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="40.0" y="40.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="49.0" y="40.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="58.0" y="40.0" width="8.0" height="7.0" rx="1"/><rect class="a-f" x="67.0" y="40.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="76.0" y="40.0" width="8.0" height="7.0" rx="1"/><rect class="a-f" x="40.0" y="48.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="49.0" y="48.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="58.0" y="48.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="67.0" y="48.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="76.0" y="48.0" width="8.0" height="7.0" rx="1"/><rect class="a-f" x="85.0" y="48.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="40.0" y="56.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="49.0" y="56.0" width="8.0" height="7.0" rx="1"/><rect class="a-f" x="58.0" y="56.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="67.0" y="56.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="76.0" y="56.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="85.0" y="56.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="94.0" y="56.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="40.0" y="64.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="49.0" y="64.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="58.0" y="64.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="67.0" y="64.0" width="8.0" height="7.0" rx="1"/><rect class="a-f" x="76.0" y="64.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="85.0" y="64.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="94.0" y="64.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="103.0" y="64.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="40.0" y="72.0" width="8.0" height="7.0" rx="1"/><rect class="a-f" x="49.0" y="72.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="58.0" y="72.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="67.0" y="72.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="76.0" y="72.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="85.0" y="72.0" width="8.0" height="7.0" rx="1"/><rect class="a-f" x="94.0" y="72.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="103.0" y="72.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="112.0" y="72.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="40.0" y="80.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="49.0" y="80.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="58.0" y="80.0" width="8.0" height="7.0" rx="1"/><rect class="a-f" x="67.0" y="80.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="76.0" y="80.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="85.0" y="80.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="94.0" y="80.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="103.0" y="80.0" width="8.0" height="7.0" rx="1"/><rect class="a-f" x="112.0" y="80.0" width="8.0" height="7.0" rx="1"/><rect class="a-g" x="121.0" y="80.0" width="8.0" height="7.0" rx="1"/><rect class="a-f" x="49.0" y="64.0" width="8.0" height="7.0" rx="1"/><rect class="a-f" x="40.0" y="24.0" width="8.0" height="7.0" rx="1"/><rect class="a-f" x="76.0" y="48.0" width="8.0" height="7.0" rx="1"/></svg></span><span class="al-meta"><span class="al-tag">Language models</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Attention: how a transformer decides what to look at</span><span class="al-d">See self-attention weights for a sentence: queries, keys, softmax and temperature, three heads and a causal mask, all interactive.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/sampling-temperature-top-p" data-tag="language models" data-q="sampling: temperature, top-k and top-p sampling"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-f" x="18.0" y="16.0" width="9.0" height="62.0" rx="1"/><rect class="a-f" x="31.0" y="32.0" width="9.0" height="46.0" rx="1"/><rect class="a-f" x="44.0" y="44.0" width="9.0" height="34.0" rx="1"/><rect class="a-f" x="57.0" y="54.0" width="9.0" height="24.0" rx="1"/><rect class="a-g" x="70.0" y="63.0" width="9.0" height="15.0" rx="1"/><rect class="a-g" x="83.0" y="68.0" width="9.0" height="10.0" rx="1"/><rect class="a-g" x="96.0" y="72.0" width="9.0" height="6.0" rx="1"/><rect class="a-g" x="109.0" y="75.0" width="9.0" height="3.0" rx="1"/><rect class="a-g" x="122.0" y="76.0" width="9.0" height="2.0" rx="1"/><rect class="a-g" x="135.0" y="77.0" width="9.0" height="1.0" rx="1"/><path class="a-l2" d="M10.0 78.0L150.0 78.0"/><path class="a-b" d="M70.0 10.0L70.0 80.0"/></svg></span><span class="al-meta"><span class="al-tag">Language models</span><span>Beginner</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Sampling: temperature, top-k and top-p</span><span class="al-d">See how a language model picks the next word: temperature, top-k and top-p reshape the probabilities, and 200 samples show what comes out.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/positional-encoding" data-tag="language models" data-q="positional encoding: sinusoids and rotary embeddings (rope) positional encoding"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><path class="a-l" d="M10.0 14.0 L11.4 15.6 L12.8 16.9 L14.2 17.5 L15.6 17.2 L17.0 16.2 L18.4 14.7 L19.8 13.0 L21.2 11.5 L22.6 10.7 L24.0 10.6 L25.4 11.3 L26.8 12.6 L28.2 14.3 L29.6 15.9 L31.0 17.1 L32.4 17.5 L33.8 17.1 L35.2 16.0 L36.6 14.4 L38.0 12.7 L39.4 11.3 L40.8 10.6 L42.2 10.6 L43.6 11.5 L45.0 12.9 L46.4 14.6 L47.8 16.2 L49.2 17.2 L50.6 17.5 L52.0 17.0 L53.4 15.7 L54.8 14.1 L56.2 12.4 L57.6 11.1 L59.0 10.5 L60.4 10.7 L61.8 11.7 L63.2 13.2 L64.6 14.9 L66.0 16.4 L67.4 17.3 L68.8 17.5 L70.2 16.8 L71.6 15.5 L73.0 13.8 L74.4 12.2 L75.8 11.0 L77.2 10.5 L78.6 10.8 L80.0 11.9 L81.4 13.5 L82.8 15.2 L84.2 16.6 L85.6 17.4 L87.0 17.4 L88.4 16.6 L89.8 15.2 L91.2 13.5 L92.6 11.9 L94.0 10.8 L95.4 10.5 L96.8 11.0 L98.2 12.2 L99.6 13.8 L101.0 15.5 L102.4 16.8 L103.8 17.5 L105.2 17.3 L106.6 16.4 L108.0 14.9 L109.4 13.2 L110.8 11.7 L112.2 10.7 L113.6 10.5 L115.0 11.2 L116.4 12.4 L117.8 14.1 L119.2 15.7 L120.6 17.0 L122.0 17.5 L123.4 17.2 L124.8 16.1 L126.2 14.6 L127.6 12.9 L129.0 11.5 L130.4 10.6 L131.8 10.6 L133.2 11.3 L134.6 12.7 L136.0 14.4 L137.4 16.0 L138.8 17.1 L140.2 17.5 L141.6 17.1 L143.0 15.9 L144.4 14.3 L145.8 12.6 L147.2 11.3 L148.6 10.6 L150.0 10.7"/><path class="a-l" d="M10.0 23.0 L11.4 24.1 L12.8 25.0 L14.2 25.8 L15.6 26.3 L17.0 26.5 L18.4 26.4 L19.8 25.9 L21.2 25.2 L22.6 24.3 L24.0 23.3 L25.4 22.2 L26.8 21.2 L28.2 20.4 L29.6 19.8 L31.0 19.5 L32.4 19.6 L33.8 19.9 L35.2 20.6 L36.6 21.4 L38.0 22.4 L39.4 23.5 L40.8 24.5 L42.2 25.4 L43.6 26.1 L45.0 26.4 L46.4 26.5 L47.8 26.2 L49.2 25.6 L50.6 24.8 L52.0 23.8 L53.4 22.8 L54.8 21.7 L56.2 20.8 L57.6 20.1 L59.0 19.6 L60.4 19.5 L61.8 19.7 L63.2 20.2 L64.6 21.0 L66.0 21.9 L67.4 23.0 L68.8 24.0 L70.2 25.0 L71.6 25.8 L73.0 26.3 L74.4 26.5 L75.8 26.4 L77.2 26.0 L78.6 25.3 L80.0 24.3 L81.4 23.3 L82.8 22.2 L84.2 21.3 L85.6 20.4 L87.0 19.8 L88.4 19.5 L89.8 19.6 L91.2 19.9 L92.6 20.5 L94.0 21.4 L95.4 22.4 L96.8 23.5 L98.2 24.5 L99.6 25.4 L101.0 26.0 L102.4 26.4 L103.8 26.5 L105.2 26.2 L106.6 25.7 L108.0 24.8 L109.4 23.9 L110.8 22.8 L112.2 21.8 L113.6 20.8 L115.0 20.1 L116.4 19.6 L117.8 19.5 L119.2 19.7 L120.6 20.2 L122.0 20.9 L123.4 21.9 L124.8 22.9 L126.2 24.0 L127.6 25.0 L129.0 25.7 L130.4 26.3 L131.8 26.5 L133.2 26.4 L134.6 26.0 L136.0 25.3 L137.4 24.4 L138.8 23.3 L140.2 22.3 L141.6 21.3 L143.0 20.4 L144.4 19.8 L145.8 19.5 L147.2 19.5 L148.6 19.9 L150.0 20.5"/><path class="a-b" d="M10.0 32.0 L11.4 32.7 L12.8 33.3 L14.2 33.9 L15.6 34.4 L17.0 34.9 L18.4 35.2 L19.8 35.4 L21.2 35.5 L22.6 35.5 L24.0 35.3 L25.4 35.0 L26.8 34.6 L28.2 34.1 L29.6 33.6 L31.0 32.9 L32.4 32.3 L33.8 31.6 L35.2 31.0 L36.6 30.3 L38.0 29.8 L39.4 29.3 L40.8 28.9 L42.2 28.7 L43.6 28.5 L45.0 28.5 L46.4 28.6 L47.8 28.9 L49.2 29.2 L50.6 29.7 L52.0 30.2 L53.4 30.8 L54.8 31.4 L56.2 32.1 L57.6 32.8 L59.0 33.4 L60.4 34.0 L61.8 34.5 L63.2 34.9 L64.6 35.2 L66.0 35.4 L67.4 35.5 L68.8 35.4 L70.2 35.3 L71.6 35.0 L73.0 34.5 L74.4 34.0 L75.8 33.5 L77.2 32.8 L78.6 32.2 L80.0 31.5 L81.4 30.8 L82.8 30.2 L84.2 29.7 L85.6 29.2 L87.0 28.9 L88.4 28.6 L89.8 28.5 L91.2 28.5 L92.6 28.7 L94.0 28.9 L95.4 29.3 L96.8 29.7 L98.2 30.3 L99.6 30.9 L101.0 31.6 L102.4 32.2 L103.8 32.9 L105.2 33.5 L106.6 34.1 L108.0 34.6 L109.4 35.0 L110.8 35.3 L112.2 35.5 L113.6 35.5 L115.0 35.4 L116.4 35.2 L117.8 34.9 L119.2 34.5 L120.6 33.9 L122.0 33.3 L123.4 32.7 L124.8 32.0 L126.2 31.4 L127.6 30.7 L129.0 30.1 L130.4 29.6 L131.8 29.2 L133.2 28.8 L134.6 28.6 L136.0 28.5 L137.4 28.5 L138.8 28.7 L140.2 29.0 L141.6 29.4 L143.0 29.8 L144.4 30.4 L145.8 31.0 L147.2 31.7 L148.6 32.3 L150.0 33.0"/><path class="a-l" d="M10.0 41.0 L11.4 41.4 L12.8 41.8 L14.2 42.2 L15.6 42.6 L17.0 43.0 L18.4 43.3 L19.8 43.6 L21.2 43.9 L22.6 44.1 L24.0 44.3 L25.4 44.4 L26.8 44.5 L28.2 44.5 L29.6 44.5 L31.0 44.4 L32.4 44.3 L33.8 44.1 L35.2 43.9 L36.6 43.7 L38.0 43.4 L39.4 43.1 L40.8 42.7 L42.2 42.3 L43.6 41.9 L45.0 41.5 L46.4 41.1 L47.8 40.7 L49.2 40.3 L50.6 39.9 L52.0 39.5 L53.4 39.1 L54.8 38.8 L56.2 38.5 L57.6 38.2 L59.0 38.0 L60.4 37.8 L61.8 37.6 L63.2 37.5 L64.6 37.5 L66.0 37.5 L67.4 37.6 L68.8 37.7 L70.2 37.8 L71.6 38.0 L73.0 38.3 L74.4 38.5 L75.8 38.9 L77.2 39.2 L78.6 39.6 L80.0 40.0 L81.4 40.4 L82.8 40.8 L84.2 41.2 L85.6 41.6 L87.0 42.0 L88.4 42.4 L89.8 42.8 L91.2 43.1 L92.6 43.4 L94.0 43.7 L95.4 44.0 L96.8 44.2 L98.2 44.3 L99.6 44.4 L101.0 44.5 L102.4 44.5 L103.8 44.5 L105.2 44.4 L106.6 44.2 L108.0 44.0 L109.4 43.8 L110.8 43.5 L112.2 43.2 L113.6 42.9 L115.0 42.5 L116.4 42.1 L117.8 41.7 L119.2 41.3 L120.6 40.9 L122.0 40.5 L123.4 40.1 L124.8 39.7 L126.2 39.3 L127.6 39.0 L129.0 38.6 L130.4 38.3 L131.8 38.1 L133.2 37.9 L134.6 37.7 L136.0 37.6 L137.4 37.5 L138.8 37.5 L140.2 37.5 L141.6 37.6 L143.0 37.7 L144.4 37.9 L145.8 38.1 L147.2 38.4 L148.6 38.7 L150.0 39.0"/><path class="a-l2" d="M10.0 50.0 L11.4 50.3 L12.8 50.5 L14.2 50.8 L15.6 51.0 L17.0 51.3 L18.4 51.5 L19.8 51.7 L21.2 52.0 L22.6 52.2 L24.0 52.4 L25.4 52.6 L26.8 52.7 L28.2 52.9 L29.6 53.0 L31.0 53.2 L32.4 53.3 L33.8 53.3 L35.2 53.4 L36.6 53.5 L38.0 53.5 L39.4 53.5 L40.8 53.5 L42.2 53.5 L43.6 53.4 L45.0 53.3 L46.4 53.3 L47.8 53.2 L49.2 53.0 L50.6 52.9 L52.0 52.7 L53.4 52.6 L54.8 52.4 L56.2 52.2 L57.6 52.0 L59.0 51.8 L60.4 51.5 L61.8 51.3 L63.2 51.0 L64.6 50.8 L66.0 50.5 L67.4 50.3 L68.8 50.0 L70.2 49.7 L71.6 49.5 L73.0 49.2 L74.4 49.0 L75.8 48.7 L77.2 48.5 L78.6 48.3 L80.0 48.0 L81.4 47.8 L82.8 47.6 L84.2 47.4 L85.6 47.3 L87.0 47.1 L88.4 47.0 L89.8 46.8 L91.2 46.7 L92.6 46.7 L94.0 46.6 L95.4 46.5 L96.8 46.5 L98.2 46.5 L99.6 46.5 L101.0 46.5 L102.4 46.6 L103.8 46.7 L105.2 46.7 L106.6 46.8 L108.0 47.0 L109.4 47.1 L110.8 47.3 L112.2 47.4 L113.6 47.6 L115.0 47.8 L116.4 48.0 L117.8 48.2 L119.2 48.5 L120.6 48.7 L122.0 49.0 L123.4 49.2 L124.8 49.5 L126.2 49.7 L127.6 50.0 L129.0 50.3 L130.4 50.5 L131.8 50.8 L133.2 51.0 L134.6 51.3 L136.0 51.5 L137.4 51.7 L138.8 52.0 L140.2 52.2 L141.6 52.4 L143.0 52.6 L144.4 52.7 L145.8 52.9 L147.2 53.0 L148.6 53.1 L150.0 53.3"/><path class="a-l2" d="M10.0 59.0 L11.4 59.2 L12.8 59.3 L14.2 59.5 L15.6 59.7 L17.0 59.8 L18.4 60.0 L19.8 60.1 L21.2 60.3 L22.6 60.4 L24.0 60.6 L25.4 60.7 L26.8 60.9 L28.2 61.0 L29.6 61.1 L31.0 61.3 L32.4 61.4 L33.8 61.5 L35.2 61.6 L36.6 61.7 L38.0 61.8 L39.4 61.9 L40.8 62.0 L42.2 62.1 L43.6 62.2 L45.0 62.2 L46.4 62.3 L47.8 62.3 L49.2 62.4 L50.6 62.4 L52.0 62.5 L53.4 62.5 L54.8 62.5 L56.2 62.5 L57.6 62.5 L59.0 62.5 L60.4 62.5 L61.8 62.5 L63.2 62.4 L64.6 62.4 L66.0 62.3 L67.4 62.3 L68.8 62.2 L70.2 62.2 L71.6 62.1 L73.0 62.0 L74.4 61.9 L75.8 61.8 L77.2 61.7 L78.6 61.6 L80.0 61.5 L81.4 61.4 L82.8 61.3 L84.2 61.2 L85.6 61.0 L87.0 60.9 L88.4 60.8 L89.8 60.6 L91.2 60.5 L92.6 60.3 L94.0 60.2 L95.4 60.0 L96.8 59.8 L98.2 59.7 L99.6 59.5 L101.0 59.4 L102.4 59.2 L103.8 59.0 L105.2 58.9 L106.6 58.7 L108.0 58.5 L109.4 58.4 L110.8 58.2 L112.2 58.1 L113.6 57.9 L115.0 57.8 L116.4 57.6 L117.8 57.5 L119.2 57.3 L120.6 57.2 L122.0 57.0 L123.4 56.9 L124.8 56.8 L126.2 56.6 L127.6 56.5 L129.0 56.4 L130.4 56.3 L131.8 56.2 L133.2 56.1 L134.6 56.0 L136.0 55.9 L137.4 55.9 L138.8 55.8 L140.2 55.7 L141.6 55.7 L143.0 55.6 L144.4 55.6 L145.8 55.6 L147.2 55.5 L148.6 55.5 L150.0 55.5"/><path class="a-l2" d="M10.0 68.0 L11.4 68.1 L12.8 68.2 L14.2 68.3 L15.6 68.4 L17.0 68.5 L18.4 68.6 L19.8 68.7 L21.2 68.8 L22.6 68.9 L24.0 69.0 L25.4 69.1 L26.8 69.2 L28.2 69.3 L29.6 69.4 L31.0 69.5 L32.4 69.6 L33.8 69.7 L35.2 69.8 L36.6 69.8 L38.0 69.9 L39.4 70.0 L40.8 70.1 L42.2 70.2 L43.6 70.3 L45.0 70.3 L46.4 70.4 L47.8 70.5 L49.2 70.6 L50.6 70.6 L52.0 70.7 L53.4 70.8 L54.8 70.8 L56.2 70.9 L57.6 70.9 L59.0 71.0 L60.4 71.0 L61.8 71.1 L63.2 71.1 L64.6 71.2 L66.0 71.2 L67.4 71.3 L68.8 71.3 L70.2 71.3 L71.6 71.4 L73.0 71.4 L74.4 71.4 L75.8 71.4 L77.2 71.5 L78.6 71.5 L80.0 71.5 L81.4 71.5 L82.8 71.5 L84.2 71.5 L85.6 71.5 L87.0 71.5 L88.4 71.5 L89.8 71.5 L91.2 71.5 L92.6 71.5 L94.0 71.4 L95.4 71.4 L96.8 71.4 L98.2 71.4 L99.6 71.3 L101.0 71.3 L102.4 71.3 L103.8 71.2 L105.2 71.2 L106.6 71.2 L108.0 71.1 L109.4 71.1 L110.8 71.0 L112.2 71.0 L113.6 70.9 L115.0 70.8 L116.4 70.8 L117.8 70.7 L119.2 70.7 L120.6 70.6 L122.0 70.5 L123.4 70.5 L124.8 70.4 L126.2 70.3 L127.6 70.2 L129.0 70.1 L130.4 70.1 L131.8 70.0 L133.2 69.9 L134.6 69.8 L136.0 69.7 L137.4 69.6 L138.8 69.5 L140.2 69.4 L141.6 69.4 L143.0 69.3 L144.4 69.2 L145.8 69.1 L147.2 69.0 L148.6 68.9 L150.0 68.8"/><path class="a-l2" d="M10.0 77.0 L11.4 77.1 L12.8 77.1 L14.2 77.2 L15.6 77.3 L17.0 77.3 L18.4 77.4 L19.8 77.4 L21.2 77.5 L22.6 77.6 L24.0 77.6 L25.4 77.7 L26.8 77.8 L28.2 77.8 L29.6 77.9 L31.0 77.9 L32.4 78.0 L33.8 78.1 L35.2 78.1 L36.6 78.2 L38.0 78.2 L39.4 78.3 L40.8 78.4 L42.2 78.4 L43.6 78.5 L45.0 78.5 L46.4 78.6 L47.8 78.7 L49.2 78.7 L50.6 78.8 L52.0 78.8 L53.4 78.9 L54.8 78.9 L56.2 79.0 L57.6 79.0 L59.0 79.1 L60.4 79.1 L61.8 79.2 L63.2 79.2 L64.6 79.3 L66.0 79.3 L67.4 79.4 L68.8 79.4 L70.2 79.5 L71.6 79.5 L73.0 79.6 L74.4 79.6 L75.8 79.6 L77.2 79.7 L78.6 79.7 L80.0 79.8 L81.4 79.8 L82.8 79.8 L84.2 79.9 L85.6 79.9 L87.0 80.0 L88.4 80.0 L89.8 80.0 L91.2 80.1 L92.6 80.1 L94.0 80.1 L95.4 80.1 L96.8 80.2 L98.2 80.2 L99.6 80.2 L101.0 80.2 L102.4 80.3 L103.8 80.3 L105.2 80.3 L106.6 80.3 L108.0 80.4 L109.4 80.4 L110.8 80.4 L112.2 80.4 L113.6 80.4 L115.0 80.4 L116.4 80.4 L117.8 80.5 L119.2 80.5 L120.6 80.5 L122.0 80.5 L123.4 80.5 L124.8 80.5 L126.2 80.5 L127.6 80.5 L129.0 80.5 L130.4 80.5 L131.8 80.5 L133.2 80.5 L134.6 80.5 L136.0 80.5 L137.4 80.5 L138.8 80.5 L140.2 80.5 L141.6 80.5 L143.0 80.5 L144.4 80.4 L145.8 80.4 L147.2 80.4 L148.6 80.4 L150.0 80.4"/></svg></span><span class="al-meta"><span class="al-tag">Language models</span><span>Advanced</span><span class="al-real">Real computation</span><span class="al-lab">Code lab</span></span><span class="al-t">Positional encoding: sinusoids and rotary embeddings (RoPE)</span><span class="al-d">See how transformers know word order: the sinusoidal encoding matrix, its similarity by distance, and RoPE rotating queries and keys so scores depend on m − n.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/kv-cache" data-tag="language models" data-q="the kv cache: why generation needs it and what it costs kv cache"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-g" x="12.0" y="26.0" width="7.2" height="14.0" rx="1"/><rect class="a-g" x="12.0" y="46.0" width="7.2" height="14.0" rx="1"/><rect class="a-g" x="20.6" y="26.0" width="7.2" height="14.0" rx="1"/><rect class="a-g" x="20.6" y="46.0" width="7.2" height="14.0" rx="1"/><rect class="a-g" x="29.2" y="26.0" width="7.2" height="14.0" rx="1"/><rect class="a-g" x="29.2" y="46.0" width="7.2" height="14.0" rx="1"/><rect class="a-g" x="37.8" y="26.0" width="7.2" height="14.0" rx="1"/><rect class="a-g" x="37.8" y="46.0" width="7.2" height="14.0" rx="1"/><rect class="a-g" x="46.4" y="26.0" width="7.2" height="14.0" rx="1"/><rect class="a-g" x="46.4" y="46.0" width="7.2" height="14.0" rx="1"/><rect class="a-g" x="55.0" y="26.0" width="7.2" height="14.0" rx="1"/><rect class="a-g" x="55.0" y="46.0" width="7.2" height="14.0" rx="1"/><rect class="a-f" x="63.6" y="26.0" width="7.2" height="14.0" rx="1"/><rect class="a-f" x="63.6" y="46.0" width="7.2" height="14.0" rx="1"/><rect class="a-f" x="72.2" y="26.0" width="7.2" height="14.0" rx="1"/><rect class="a-f" x="72.2" y="46.0" width="7.2" height="14.0" rx="1"/><rect class="a-f" x="80.8" y="26.0" width="7.2" height="14.0" rx="1"/><rect class="a-f" x="80.8" y="46.0" width="7.2" height="14.0" rx="1"/><rect class="a-f" x="89.4" y="26.0" width="7.2" height="14.0" rx="1"/><rect class="a-f" x="89.4" y="46.0" width="7.2" height="14.0" rx="1"/><rect class="a-f" x="98.0" y="26.0" width="7.2" height="14.0" rx="1"/><rect class="a-f" x="98.0" y="46.0" width="7.2" height="14.0" rx="1"/><rect class="a-f" x="106.6" y="26.0" width="7.2" height="14.0" rx="1"/><rect class="a-f" x="106.6" y="46.0" width="7.2" height="14.0" rx="1"/><rect class="a-f" x="115.2" y="26.0" width="7.2" height="14.0" rx="1"/><rect class="a-f" x="115.2" y="46.0" width="7.2" height="14.0" rx="1"/><rect class="a-f" x="123.8" y="26.0" width="7.2" height="14.0" rx="1"/><rect class="a-f" x="123.8" y="46.0" width="7.2" height="14.0" rx="1"/><rect class="a-f" x="132.4" y="26.0" width="7.2" height="14.0" rx="1"/><rect class="a-f" x="132.4" y="46.0" width="7.2" height="14.0" rx="1"/><rect class="a-g" x="141.0" y="26.0" width="7.2" height="14.0" rx="1"/><rect class="a-g" x="141.0" y="46.0" width="7.2" height="14.0" rx="1"/><rect class="a-b" x="132.4" y="26.0" width="7.2" height="34.0" rx="1"/><path class="a-l2" d="M12.0 72.0L148.0 72.0"/></svg></span><span class="al-meta"><span class="al-tag">Language models</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-lab">Code lab</span></span><span class="al-t">The KV cache: why generation needs it and what it costs</span><span class="al-d">Size the KV cache of 1B, 8B and 70B-class models across context length and batch, compare MHA, GQA and MQA, and watch decoding reuse cached keys and values.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/beam-search" data-tag="language models" data-q="greedy decoding vs beam search, with length penalty beam search"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><path class="a-l2" d="M20.0 45.0L60.0 20.0"/><path class="a-l2" d="M20.0 45.0L60.0 45.0"/><path class="a-l2" d="M20.0 45.0L60.0 70.0"/><path class="a-b" d="M20.0 45.0L60.0 45.0"/><path class="a-b" d="M60.0 45.0L100.0 34.0"/><path class="a-l2" d="M60.0 45.0L100.0 52.0"/><path class="a-l2" d="M60.0 45.0L100.0 62.0"/><path class="a-l2" d="M60.0 20.0L100.0 16.0"/><path class="a-b" d="M100.0 34.0L140.0 30.0"/><path class="a-l2" d="M100.0 34.0L140.0 44.0"/><circle class="a-db" cx="20.0" cy="45.0" r="1.8"/><circle class="a-d" cx="60.0" cy="20.0" r="1.8"/><circle class="a-db" cx="60.0" cy="45.0" r="1.8"/><circle class="a-d" cx="60.0" cy="70.0" r="1.8"/><circle class="a-d" cx="100.0" cy="16.0" r="1.8"/><circle class="a-db" cx="100.0" cy="34.0" r="1.8"/><circle class="a-d" cx="100.0" cy="52.0" r="1.8"/><circle class="a-d" cx="100.0" cy="62.0" r="1.8"/><circle class="a-db" cx="140.0" cy="30.0" r="1.8"/><circle class="a-d" cx="140.0" cy="44.0" r="1.8"/></svg></span><span class="al-meta"><span class="al-tag">Language models</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-lab">Code lab</span></span><span class="al-t">Greedy decoding vs beam search, with length penalty</span><span class="al-d">Watch greedy decoding and beam search expand a small language model step by step, find a more probable sentence, and change with the length penalty.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/lora" data-tag="language models" data-q="lora: fine-tuning with a low-rank update lora"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-g" x="14.0" y="14.0" width="62.0" height="62.0" rx="2"/><path class="a-l2" d="M14.0 14.0L76.0 14.0"/><path class="a-l2" d="M14.0 21.8L76.0 21.8"/><path class="a-l2" d="M14.0 29.5L76.0 29.5"/><path class="a-l2" d="M14.0 37.2L76.0 37.2"/><path class="a-l2" d="M14.0 45.0L76.0 45.0"/><path class="a-l2" d="M14.0 52.8L76.0 52.8"/><path class="a-l2" d="M14.0 60.5L76.0 60.5"/><path class="a-l2" d="M14.0 68.2L76.0 68.2"/><path class="a-l2" d="M14.0 76.0L76.0 76.0"/><text class="a-t" x="36.0" y="48.0">W frozen</text><rect class="a-f" x="92.0" y="14.0" width="10.0" height="62.0" rx="1"/><rect class="a-f" x="108.0" y="14.0" width="40.0" height="10.0" rx="1"/><text class="a-t" x="93.0" y="82.0">B</text><text class="a-t" x="124.0" y="32.0">A</text><path class="a-b" d="M78.0 45.0L90.0 45.0"/></svg></span><span class="al-meta"><span class="al-tag">Language models</span><span>Advanced</span><span class="al-real">Real computation</span><span class="al-lab">Code lab</span></span><span class="al-t">LoRA: fine-tuning with a low-rank update</span><span class="al-d">Train LoRA&#x27;s A and B matrices to fit an update of known rank, compare with the SVD optimum, and count trainable parameters for real model sizes.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/quantization" data-tag="language models" data-q="quantisation: 8, 4 and 2 bits, scales, groups and outliers quantisation"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><path class="a-l2" d="M20.0 12.0L20.0 78.0"/><path class="a-l2" d="M37.1 12.0L37.1 78.0"/><path class="a-l2" d="M54.2 12.0L54.2 78.0"/><path class="a-l2" d="M71.3 12.0L71.3 78.0"/><path class="a-l2" d="M88.4 12.0L88.4 78.0"/><path class="a-l2" d="M105.5 12.0L105.5 78.0"/><path class="a-l2" d="M122.6 12.0L122.6 78.0"/><path class="a-l2" d="M139.7 12.0L139.7 78.0"/><path class="a-l" d="M10.0 78.0 L11.4 78.0 L12.8 78.0 L14.2 78.0 L15.6 78.0 L17.0 78.0 L18.4 78.0 L19.8 78.0 L21.2 77.9 L22.6 77.9 L24.0 77.9 L25.4 77.8 L26.8 77.8 L28.2 77.7 L29.6 77.6 L31.0 77.5 L32.4 77.4 L33.8 77.2 L35.2 76.9 L36.6 76.6 L38.0 76.3 L39.4 75.8 L40.8 75.3 L42.2 74.6 L43.6 73.9 L45.0 73.0 L46.4 71.9 L47.8 70.7 L49.2 69.2 L50.6 67.6 L52.0 65.8 L53.4 63.8 L54.8 61.6 L56.2 59.2 L57.6 56.7 L59.0 53.9 L60.4 51.0 L61.8 48.0 L63.2 45.0 L64.6 41.8 L66.0 38.8 L67.4 35.7 L68.8 32.8 L70.2 30.1 L71.6 27.6 L73.0 25.4 L74.4 23.5 L75.8 22.0 L77.2 20.9 L78.6 20.2 L80.0 20.0 L81.4 20.2 L82.8 20.9 L84.2 22.0 L85.6 23.5 L87.0 25.4 L88.4 27.6 L89.8 30.1 L91.2 32.8 L92.6 35.7 L94.0 38.8 L95.4 41.8 L96.8 45.0 L98.2 48.0 L99.6 51.0 L101.0 53.9 L102.4 56.7 L103.8 59.2 L105.2 61.6 L106.6 63.8 L108.0 65.8 L109.4 67.6 L110.8 69.2 L112.2 70.7 L113.6 71.9 L115.0 73.0 L116.4 73.9 L117.8 74.6 L119.2 75.3 L120.6 75.8 L122.0 76.3 L123.4 76.6 L124.8 76.9 L126.2 77.2 L127.6 77.4 L129.0 77.5 L130.4 77.6 L131.8 77.7 L133.2 77.8 L134.6 77.8 L136.0 77.9 L137.4 77.9 L138.8 77.9 L140.2 78.0 L141.6 78.0 L143.0 78.0 L144.4 78.0 L145.8 78.0 L147.2 78.0 L148.6 78.0 L150.0 78.0"/><rect class="a-f" x="18.0" y="78.0" width="4.0" height="0.0" rx="1"/><rect class="a-f" x="35.1" y="77.8" width="4.0" height="0.2" rx="1"/><rect class="a-f" x="52.2" y="73.5" width="4.0" height="4.5" rx="1"/><rect class="a-f" x="69.3" y="49.1" width="4.0" height="28.9" rx="1"/><rect class="a-f" x="86.4" y="20.3" width="4.0" height="57.7" rx="1"/><rect class="a-f" x="103.5" y="42.1" width="4.0" height="35.9" rx="1"/><rect class="a-f" x="120.6" y="71.0" width="4.0" height="7.0" rx="1"/><rect class="a-f" x="137.7" y="77.6" width="4.0" height="0.4" rx="1"/></svg></span><span class="al-meta"><span class="al-tag">Language models</span><span>Advanced</span><span class="al-real">Real computation</span><span class="al-lab">Code lab</span></span><span class="al-t">Quantisation: 8, 4 and 2 bits, scales, groups and outliers</span><span class="al-d">Quantise a real weight matrix to 8, 6, 4, 3 or 2 bits with per-tensor, per-channel or group scales and see the error, the effect of outliers and the memory.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/mixture-of-experts" data-tag="language models" data-q="mixture of experts: routing tokens with top-k gates mixture of experts"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><circle class="a-d" cx="22.0" cy="14.0" r="1.6"/><circle class="a-d" cx="22.0" cy="23.5" r="1.6"/><circle class="a-d" cx="22.0" cy="33.0" r="1.6"/><circle class="a-d" cx="22.0" cy="42.5" r="1.6"/><circle class="a-d" cx="22.0" cy="52.0" r="1.6"/><circle class="a-d" cx="22.0" cy="61.5" r="1.6"/><circle class="a-d" cx="22.0" cy="71.0" r="1.6"/><circle class="a-d" cx="22.0" cy="80.5" r="1.6"/><rect class="a-g" x="112.0" y="10.0" width="34.0" height="12.0" rx="3"/><rect class="a-f" x="112.0" y="28.0" width="34.0" height="12.0" rx="3"/><rect class="a-g" x="112.0" y="46.0" width="34.0" height="12.0" rx="3"/><rect class="a-f" x="112.0" y="64.0" width="34.0" height="12.0" rx="3"/><path class="a-l2" d="M24.0 14.0 L68.0 14.0 L68.0 34.0 L110.0 34.0"/><path class="a-b" d="M24.0 23.5 L68.0 23.5 L68.0 70.0 L110.0 70.0"/><path class="a-l2" d="M24.0 33.0 L68.0 33.0 L68.0 16.0 L110.0 16.0"/><path class="a-b" d="M24.0 42.5 L68.0 42.5 L68.0 34.0 L110.0 34.0"/><path class="a-l2" d="M24.0 52.0 L68.0 52.0 L68.0 52.0 L110.0 52.0"/><path class="a-b" d="M24.0 61.5 L68.0 61.5 L68.0 70.0 L110.0 70.0"/><path class="a-l2" d="M24.0 71.0 L68.0 71.0 L68.0 34.0 L110.0 34.0"/><path class="a-b" d="M24.0 80.5 L68.0 80.5 L68.0 70.0 L110.0 70.0"/></svg></span><span class="al-meta"><span class="al-tag">Language models</span><span>Advanced</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Mixture of experts: routing tokens with top-k gates</span><span class="al-d">Type a sentence and watch a router send each token to its top-k experts, with capacity limits, dropped tokens and the load-balancing loss.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a></div></section><section class="al-month"><h2 class="al-mh" id="retrieval-(rag)">Retrieval (RAG) <span>15</span></h2><div class="al-grid"><a class="al-card" href="/ai-simulations/rag-types" data-tag="retrieval (rag)" data-q="rag types compared: naive, hybrid, rerank, hyde, corrective, agentic rag types"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><path class="a-l2" d="M22.0 45.0L123.6 54.3"/><path class="a-l2" d="M22.0 45.0L112.3 27.2"/><path class="a-l2" d="M22.0 45.0L122.5 55.7"/><path class="a-l2" d="M22.0 45.0L105.2 45.8"/><path class="a-l2" d="M22.0 45.0L131.0 52.6"/><path class="a-l2" d="M22.0 45.0L105.5 28.0"/><path class="a-l2" d="M22.0 45.0L100.2 32.0"/><path class="a-l2" d="M22.0 45.0L96.7 42.6"/><path class="a-l2" d="M22.0 45.0L124.1 15.9"/><path class="a-l2" d="M22.0 45.0L110.9 16.6"/><path class="a-l2" d="M22.0 45.0L145.8 42.7"/><path class="a-l2" d="M22.0 45.0L129.6 64.3"/><path class="a-l2" d="M22.0 45.0L119.2 67.6"/><path class="a-l2" d="M22.0 45.0L141.2 58.7"/><path class="a-l2" d="M22.0 45.0L133.1 14.8"/><path class="a-l2" d="M22.0 45.0L103.7 13.0"/><path class="a-l2" d="M22.0 45.0L100.0 65.4"/><path class="a-l2" d="M22.0 45.0L149.0 75.7"/><path class="a-l2" d="M22.0 45.0L149.9 77.7"/><path class="a-b" d="M22.0 45.0L119.9 48.9"/><path class="a-b" d="M22.0 45.0L120.2 30.4"/><path class="a-b" d="M22.0 45.0L122.9 50.8"/><circle class="a-d" cx="119.9" cy="48.9" r="1.1"/><circle class="a-d" cx="145.8" cy="42.7" r="1.1"/><circle class="a-d" cx="122.9" cy="50.8" r="1.1"/><circle class="a-d" cx="105.2" cy="45.8" r="1.1"/><circle class="a-d" cx="129.6" cy="64.3" r="1.1"/><circle class="a-d" cx="100.2" cy="32.0" r="1.1"/><circle class="a-d" cx="100.0" cy="65.4" r="1.1"/><circle class="a-d" cx="133.1" cy="14.8" r="1.1"/><circle class="a-d" cx="149.0" cy="75.7" r="1.1"/><circle class="a-d" cx="131.0" cy="52.6" r="1.1"/><circle class="a-d" cx="103.7" cy="13.0" r="1.1"/><circle class="a-d" cx="124.1" cy="15.9" r="1.1"/><circle class="a-d" cx="105.5" cy="28.0" r="1.1"/><circle class="a-d" cx="96.7" cy="42.6" r="1.1"/><circle class="a-d" cx="119.2" cy="67.6" r="1.1"/><circle class="a-d" cx="123.6" cy="54.3" r="1.1"/><circle class="a-d" cx="122.5" cy="55.7" r="1.1"/><circle class="a-d" cx="120.2" cy="30.4" r="1.1"/><circle class="a-d" cx="149.9" cy="77.7" r="1.1"/><circle class="a-d" cx="141.2" cy="58.7" r="1.1"/><circle class="a-d" cx="112.3" cy="27.2" r="1.1"/><circle class="a-d" cx="110.9" cy="16.6" r="1.1"/><circle class="a-db" cx="119.9" cy="48.9" r="1.8"/><circle class="a-db" cx="120.2" cy="30.4" r="1.8"/><circle class="a-db" cx="122.9" cy="50.8" r="1.8"/><circle class="a-db" cx="22.0" cy="45.0" r="2.6"/></svg></span><span class="al-meta"><span class="al-tag">Retrieval (RAG)</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">RAG types compared: naive, hybrid, rerank, HyDE, corrective, agentic</span><span class="al-d">Run seven retrieval strategies on the same corpus and compare what each one finds: naive, hybrid, rerank, HyDE, query rewriting, corrective, agentic.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/rag-naive" data-tag="retrieval (rag)" data-q="naive rag: embed, search, take the top three naive rag"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><path class="a-l2" d="M22.0 45.0L122.8 27.6"/><path class="a-l2" d="M22.0 45.0L139.4 45.8"/><path class="a-l2" d="M22.0 45.0L122.4 19.6"/><path class="a-l2" d="M22.0 45.0L138.7 27.3"/><path class="a-l2" d="M22.0 45.0L137.9 25.8"/><path class="a-l2" d="M22.0 45.0L104.1 57.5"/><path class="a-l2" d="M22.0 45.0L96.9 33.7"/><path class="a-l2" d="M22.0 45.0L127.2 16.6"/><path class="a-l2" d="M22.0 45.0L95.2 36.5"/><path class="a-l2" d="M22.0 45.0L142.6 27.5"/><path class="a-l2" d="M22.0 45.0L143.1 52.6"/><path class="a-l2" d="M22.0 45.0L147.4 36.3"/><path class="a-l2" d="M22.0 45.0L106.9 64.3"/><path class="a-l2" d="M22.0 45.0L107.8 14.8"/><path class="a-l2" d="M22.0 45.0L105.3 14.5"/><path class="a-l2" d="M22.0 45.0L103.5 66.0"/><path class="a-l2" d="M22.0 45.0L149.9 60.8"/><path class="a-l2" d="M22.0 45.0L101.9 73.2"/><path class="a-l2" d="M22.0 45.0L148.3 71.0"/><path class="a-b" d="M22.0 45.0L113.9 44.0"/><path class="a-b" d="M22.0 45.0L119.9 48.4"/><path class="a-b" d="M22.0 45.0L129.9 43.6"/><circle class="a-d" cx="104.1" cy="57.5" r="1.1"/><circle class="a-d" cx="129.9" cy="43.6" r="1.1"/><circle class="a-d" cx="106.9" cy="64.3" r="1.1"/><circle class="a-d" cx="139.4" cy="45.8" r="1.1"/><circle class="a-d" cx="122.8" cy="27.6" r="1.1"/><circle class="a-d" cx="95.2" cy="36.5" r="1.1"/><circle class="a-d" cx="127.2" cy="16.6" r="1.1"/><circle class="a-d" cx="138.7" cy="27.3" r="1.1"/><circle class="a-d" cx="107.8" cy="14.8" r="1.1"/><circle class="a-d" cx="149.9" cy="60.8" r="1.1"/><circle class="a-d" cx="143.1" cy="52.6" r="1.1"/><circle class="a-d" cx="96.9" cy="33.7" r="1.1"/><circle class="a-d" cx="122.4" cy="19.6" r="1.1"/><circle class="a-d" cx="147.4" cy="36.3" r="1.1"/><circle class="a-d" cx="103.5" cy="66.0" r="1.1"/><circle class="a-d" cx="101.9" cy="73.2" r="1.1"/><circle class="a-d" cx="119.9" cy="48.4" r="1.1"/><circle class="a-d" cx="113.9" cy="44.0" r="1.1"/><circle class="a-d" cx="105.3" cy="14.5" r="1.1"/><circle class="a-d" cx="142.6" cy="27.5" r="1.1"/><circle class="a-d" cx="137.9" cy="25.8" r="1.1"/><circle class="a-d" cx="148.3" cy="71.0" r="1.1"/><circle class="a-db" cx="113.9" cy="44.0" r="1.8"/><circle class="a-db" cx="119.9" cy="48.4" r="1.8"/><circle class="a-db" cx="129.9" cy="43.6" r="1.8"/><circle class="a-db" cx="22.0" cy="45.0" r="2.6"/></svg></span><span class="al-meta"><span class="al-tag">Retrieval (RAG)</span><span>Beginner</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Naive RAG: embed, search, take the top three</span><span class="al-d">The baseline of retrieval-augmented generation, step by step: embed the question, score every chunk by cosine similarity and pass the top three to the model.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/rag-hybrid" data-tag="retrieval (rag)" data-q="hybrid search: bm25 keywords plus vectors, fused by rank hybrid search"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><path class="a-l2" d="M22.0 45.0L131.7 24.2"/><path class="a-l2" d="M22.0 45.0L97.7 47.1"/><path class="a-l2" d="M22.0 45.0L141.7 50.5"/><path class="a-l2" d="M22.0 45.0L112.2 63.6"/><path class="a-l2" d="M22.0 45.0L105.1 20.0"/><path class="a-l2" d="M22.0 45.0L143.5 50.9"/><path class="a-l2" d="M22.0 45.0L143.9 27.3"/><path class="a-l2" d="M22.0 45.0L133.9 15.2"/><path class="a-l2" d="M22.0 45.0L147.8 33.1"/><path class="a-l2" d="M22.0 45.0L147.9 49.4"/><path class="a-l2" d="M22.0 45.0L147.5 55.0"/><path class="a-l2" d="M22.0 45.0L102.8 66.5"/><path class="a-l2" d="M22.0 45.0L147.7 21.3"/><path class="a-l2" d="M22.0 45.0L128.1 73.5"/><path class="a-l2" d="M22.0 45.0L146.6 63.1"/><path class="a-l2" d="M22.0 45.0L132.7 73.0"/><path class="a-l2" d="M22.0 45.0L109.2 75.1"/><path class="a-l2" d="M22.0 45.0L96.8 73.1"/><path class="a-l2" d="M22.0 45.0L96.3 77.9"/><path class="a-b" d="M22.0 45.0L130.8 34.8"/><path class="a-b" d="M22.0 45.0L112.1 24.5"/><path class="a-b" d="M22.0 45.0L133.4 52.4"/><circle class="a-d" cx="147.7" cy="21.3" r="1.1"/><circle class="a-d" cx="96.3" cy="77.9" r="1.1"/><circle class="a-d" cx="105.1" cy="20.0" r="1.1"/><circle class="a-d" cx="130.8" cy="34.8" r="1.1"/><circle class="a-d" cx="143.9" cy="27.3" r="1.1"/><circle class="a-d" cx="147.8" cy="33.1" r="1.1"/><circle class="a-d" cx="128.1" cy="73.5" r="1.1"/><circle class="a-d" cx="132.7" cy="73.0" r="1.1"/><circle class="a-d" cx="133.9" cy="15.2" r="1.1"/><circle class="a-d" cx="143.5" cy="50.9" r="1.1"/><circle class="a-d" cx="112.1" cy="24.5" r="1.1"/><circle class="a-d" cx="141.7" cy="50.5" r="1.1"/><circle class="a-d" cx="146.6" cy="63.1" r="1.1"/><circle class="a-d" cx="147.9" cy="49.4" r="1.1"/><circle class="a-d" cx="133.4" cy="52.4" r="1.1"/><circle class="a-d" cx="109.2" cy="75.1" r="1.1"/><circle class="a-d" cx="112.2" cy="63.6" r="1.1"/><circle class="a-d" cx="147.5" cy="55.0" r="1.1"/><circle class="a-d" cx="102.8" cy="66.5" r="1.1"/><circle class="a-d" cx="131.7" cy="24.2" r="1.1"/><circle class="a-d" cx="97.7" cy="47.1" r="1.1"/><circle class="a-d" cx="96.8" cy="73.1" r="1.1"/><circle class="a-db" cx="130.8" cy="34.8" r="1.8"/><circle class="a-db" cx="112.1" cy="24.5" r="1.8"/><circle class="a-db" cx="133.4" cy="52.4" r="1.8"/><circle class="a-db" cx="22.0" cy="45.0" r="2.6"/><path class="a-l" d="M22.0 45.0L141.4 42.6"/><path class="a-l" d="M22.0 45.0L102.4 49.7"/><path class="a-l" d="M22.0 45.0L147.6 41.9"/><path class="a-l" d="M22.0 45.0L108.1 43.3"/></svg></span><span class="al-meta"><span class="al-tag">Retrieval (RAG)</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Hybrid search: BM25 keywords plus vectors, fused by rank</span><span class="al-d">See hybrid retrieval at work: a vector ranking, a BM25 keyword ranking and reciprocal rank fusion that merges them without tuning score scales.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/rag-rerank" data-tag="retrieval (rag)" data-q="retrieve and rerank with a cross-encoder rerank"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-g" x="30.0" y="10.0" width="40.0" height="6.0" rx="1"/><rect class="a-g" x="30.0" y="19.0" width="37.0" height="6.0" rx="1"/><rect class="a-g" x="30.0" y="28.0" width="34.0" height="6.0" rx="1"/><rect class="a-g" x="30.0" y="37.0" width="31.0" height="6.0" rx="1"/><rect class="a-g" x="30.0" y="46.0" width="28.0" height="6.0" rx="1"/><rect class="a-g" x="30.0" y="55.0" width="25.0" height="6.0" rx="1"/><rect class="a-g" x="30.0" y="64.0" width="22.0" height="6.0" rx="1"/><rect class="a-g" x="30.0" y="73.0" width="19.0" height="6.0" rx="1"/><rect class="a-f" x="98.0" y="10.0" width="46.0" height="6.0" rx="1"/><rect class="a-f" x="98.0" y="19.0" width="41.0" height="6.0" rx="1"/><rect class="a-f" x="98.0" y="28.0" width="36.0" height="6.0" rx="1"/><rect class="a-g" x="98.0" y="37.0" width="31.0" height="6.0" rx="1"/><rect class="a-g" x="98.0" y="46.0" width="26.0" height="6.0" rx="1"/><rect class="a-g" x="98.0" y="55.0" width="21.0" height="6.0" rx="1"/><rect class="a-g" x="98.0" y="64.0" width="16.0" height="6.0" rx="1"/><rect class="a-g" x="98.0" y="73.0" width="11.0" height="6.0" rx="1"/><path class="a-b" d="M70.0 40.0 L98.0 13.0"/><path class="a-b" d="M70.0 58.0 L98.0 22.0"/></svg></span><span class="al-meta"><span class="al-tag">Retrieval (RAG)</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Retrieve and rerank with a cross-encoder</span><span class="al-d">Retrieve eight candidates fast, then rerank them with a cross-encoder that reads question and passage together, and watch the relevant chunks climb.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/rag-multi-query" data-tag="retrieval (rag)" data-q="multi-query rewriting: ask the same question several ways multi-query"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><circle class="a-db" cx="20.0" cy="24.0" r="2.2"/><path class="a-b" d="M20.0 24.0L108.0 30.0"/><circle class="a-db" cx="20.0" cy="45.0" r="2.2"/><path class="a-b" d="M20.0 45.0L120.0 42.0"/><circle class="a-db" cx="20.0" cy="66.0" r="2.2"/><path class="a-b" d="M20.0 66.0L132.0 54.0"/><circle class="a-d" cx="105.5" cy="30.7" r="0.9"/><circle class="a-d" cx="131.4" cy="19.5" r="0.9"/><circle class="a-d" cx="120.0" cy="19.9" r="0.9"/><circle class="a-d" cx="137.4" cy="44.0" r="0.9"/><circle class="a-d" cx="141.0" cy="37.1" r="0.9"/><circle class="a-d" cx="127.6" cy="39.2" r="0.9"/><circle class="a-d" cx="111.7" cy="43.4" r="0.9"/><circle class="a-d" cx="104.4" cy="38.5" r="0.9"/><circle class="a-d" cx="131.8" cy="42.6" r="0.9"/><circle class="a-d" cx="116.3" cy="63.5" r="0.9"/><circle class="a-d" cx="122.8" cy="36.2" r="0.9"/><circle class="a-d" cx="124.2" cy="36.9" r="0.9"/><circle class="a-d" cx="116.6" cy="38.6" r="0.9"/><circle class="a-d" cx="150.4" cy="42.2" r="0.9"/><circle class="a-d" cx="124.5" cy="48.6" r="0.9"/><circle class="a-d" cx="150.2" cy="39.8" r="0.9"/><circle class="a-d" cx="113.3" cy="66.3" r="0.9"/><circle class="a-d" cx="101.6" cy="38.4" r="0.9"/><circle class="a-d" cx="131.3" cy="64.4" r="0.9"/><circle class="a-d" cx="108.7" cy="18.2" r="0.9"/><circle class="a-d" cx="131.3" cy="36.9" r="0.9"/><circle class="a-d" cx="116.6" cy="46.5" r="0.9"/><circle class="a-d" cx="125.1" cy="44.9" r="0.9"/><circle class="a-d" cx="116.0" cy="54.6" r="0.9"/><circle class="a-d" cx="143.1" cy="42.3" r="0.9"/><circle class="a-d" cx="115.7" cy="49.2" r="0.9"/></svg></span><span class="al-meta"><span class="al-tag">Retrieval (RAG)</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Multi-query rewriting: ask the same question several ways</span><span class="al-d">An LLM rewrites the question into several versions in the documents&#x27; vocabulary; each searches on its own and the results are merged by rank fusion.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/rag-hyde" data-tag="retrieval (rag)" data-q="hyde: search with a hypothetical answer hyde"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><circle class="a-db" cx="16.0" cy="45.0" r="2.4"/><rect class="a-f" x="40.0" y="34.0" width="40.0" height="22.0" rx="3"/><path class="a-l" d="M46.0 40.0L74.0 40.0"/><path class="a-l" d="M46.0 45.0L74.0 45.0"/><path class="a-l" d="M46.0 50.0L74.0 50.0"/><path class="a-l" d="M18.0 45.0L40.0 45.0"/><path class="a-b" d="M80.0 45.0L118.0 30.0"/><path class="a-b" d="M80.0 45.0L122.0 52.0"/><circle class="a-d" cx="129.5" cy="65.3" r="0.9"/><circle class="a-d" cx="138.1" cy="53.4" r="0.9"/><circle class="a-d" cx="132.8" cy="47.2" r="0.9"/><circle class="a-d" cx="133.2" cy="44.0" r="0.9"/><circle class="a-d" cx="116.6" cy="36.8" r="0.9"/><circle class="a-d" cx="111.6" cy="46.8" r="0.9"/><circle class="a-d" cx="133.0" cy="58.8" r="0.9"/><circle class="a-d" cx="132.4" cy="47.3" r="0.9"/><circle class="a-d" cx="134.4" cy="44.8" r="0.9"/><circle class="a-d" cx="103.4" cy="53.9" r="0.9"/><circle class="a-d" cx="120.7" cy="46.8" r="0.9"/><circle class="a-d" cx="123.1" cy="69.9" r="0.9"/><circle class="a-d" cx="140.4" cy="48.5" r="0.9"/><circle class="a-d" cx="112.4" cy="52.6" r="0.9"/><circle class="a-d" cx="115.3" cy="54.9" r="0.9"/><circle class="a-d" cx="107.6" cy="44.2" r="0.9"/><circle class="a-d" cx="137.9" cy="37.6" r="0.9"/><circle class="a-d" cx="127.7" cy="46.4" r="0.9"/><circle class="a-d" cx="138.1" cy="40.8" r="0.9"/><circle class="a-d" cx="135.0" cy="63.0" r="0.9"/><circle class="a-d" cx="107.9" cy="37.5" r="0.9"/><circle class="a-d" cx="132.2" cy="39.3" r="0.9"/></svg></span><span class="al-meta"><span class="al-tag">Retrieval (RAG)</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">HyDE: search with a hypothetical answer</span><span class="al-d">Hypothetical Document Embeddings: an LLM drafts an answer without retrieval, and the draft, which looks like the documents, is used as the search query.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/rag-contextual" data-tag="retrieval (rag)" data-q="contextual chunks and parent–child retrieval contextual chunks"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-g" x="18.0" y="10.0" width="124.0" height="14.0" rx="2"/><rect class="a-f" x="18.0" y="10.0" width="34.0" height="14.0" rx="2"/><path class="a-l2" d="M58.0 17.0L136.0 17.0"/><rect class="a-g" x="18.0" y="28.0" width="124.0" height="14.0" rx="2"/><rect class="a-f" x="18.0" y="28.0" width="34.0" height="14.0" rx="2"/><path class="a-l2" d="M58.0 35.0L136.0 35.0"/><rect class="a-g" x="18.0" y="46.0" width="124.0" height="14.0" rx="2"/><rect class="a-f" x="18.0" y="46.0" width="34.0" height="14.0" rx="2"/><path class="a-l2" d="M58.0 53.0L136.0 53.0"/><rect class="a-g" x="18.0" y="64.0" width="124.0" height="14.0" rx="2"/><rect class="a-f" x="18.0" y="64.0" width="34.0" height="14.0" rx="2"/><path class="a-l2" d="M58.0 71.0L136.0 71.0"/></svg></span><span class="al-meta"><span class="al-tag">Retrieval (RAG)</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Contextual chunks and parent–child retrieval</span><span class="al-d">Chunks lose meaning when cut out of their document. Add a short context header before indexing and compare rankings with and without it.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/rag-corrective" data-tag="retrieval (rag)" data-q="corrective rag: grade the retrieval, search again if it is weak corrective rag"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><path class="a-l2" d="M22.0 45.0L131.0 53.0"/><path class="a-l2" d="M22.0 45.0L135.2 50.0"/><path class="a-l2" d="M22.0 45.0L126.8 57.5"/><path class="a-l2" d="M22.0 45.0L141.2 41.0"/><path class="a-l2" d="M22.0 45.0L134.8 23.1"/><path class="a-l2" d="M22.0 45.0L141.6 31.9"/><path class="a-l2" d="M22.0 45.0L97.9 33.0"/><path class="a-l2" d="M22.0 45.0L103.0 56.1"/><path class="a-l2" d="M22.0 45.0L138.0 55.7"/><path class="a-l2" d="M22.0 45.0L101.9 56.2"/><path class="a-l2" d="M22.0 45.0L103.5 21.9"/><path class="a-l2" d="M22.0 45.0L134.9 60.3"/><path class="a-l2" d="M22.0 45.0L141.3 26.2"/><path class="a-l2" d="M22.0 45.0L98.3 63.8"/><path class="a-l2" d="M22.0 45.0L105.0 70.6"/><path class="a-l2" d="M22.0 45.0L104.1 72.3"/><path class="a-l2" d="M22.0 45.0L101.8 72.3"/><path class="a-l2" d="M22.0 45.0L96.8 69.2"/><path class="a-l2" d="M22.0 45.0L132.9 76.7"/><path class="a-b" d="M22.0 45.0L125.4 37.3"/><path class="a-b" d="M22.0 45.0L116.0 27.2"/><path class="a-b" d="M22.0 45.0L129.1 52.1"/><circle class="a-d" cx="116.0" cy="27.2" r="1.1"/><circle class="a-d" cx="104.1" cy="72.3" r="1.1"/><circle class="a-d" cx="126.8" cy="57.5" r="1.1"/><circle class="a-d" cx="125.4" cy="37.3" r="1.1"/><circle class="a-d" cx="135.2" cy="50.0" r="1.1"/><circle class="a-d" cx="138.0" cy="55.7" r="1.1"/><circle class="a-d" cx="141.2" cy="41.0" r="1.1"/><circle class="a-d" cx="103.5" cy="21.9" r="1.1"/><circle class="a-d" cx="134.8" cy="23.1" r="1.1"/><circle class="a-d" cx="131.0" cy="53.0" r="1.1"/><circle class="a-d" cx="98.3" cy="63.8" r="1.1"/><circle class="a-d" cx="96.8" cy="69.2" r="1.1"/><circle class="a-d" cx="141.3" cy="26.2" r="1.1"/><circle class="a-d" cx="134.9" cy="60.3" r="1.1"/><circle class="a-d" cx="101.9" cy="56.2" r="1.1"/><circle class="a-d" cx="101.8" cy="72.3" r="1.1"/><circle class="a-d" cx="132.9" cy="76.7" r="1.1"/><circle class="a-d" cx="97.9" cy="33.0" r="1.1"/><circle class="a-d" cx="129.1" cy="52.1" r="1.1"/><circle class="a-d" cx="105.0" cy="70.6" r="1.1"/><circle class="a-d" cx="141.6" cy="31.9" r="1.1"/><circle class="a-d" cx="103.0" cy="56.1" r="1.1"/><circle class="a-db" cx="125.4" cy="37.3" r="1.8"/><circle class="a-db" cx="116.0" cy="27.2" r="1.8"/><circle class="a-db" cx="129.1" cy="52.1" r="1.8"/><circle class="a-db" cx="22.0" cy="45.0" r="2.6"/><path class="a-b" d="M60 70l6 6 12-14"/><path class="a-l" d="M60 20l12 12M72 20l-12 12"/></svg></span><span class="al-meta"><span class="al-tag">Retrieval (RAG)</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Corrective RAG: grade the retrieval, search again if it is weak</span><span class="al-d">Corrective RAG grades every retrieved chunk as correct, ambiguous or incorrect, rewrites and searches again when the evidence is weak, then strips noise.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/rag-self" data-tag="retrieval (rag)" data-q="self-rag: the model decides when to retrieve and critiques itself self-rag"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-g" x="12.0" y="36.0" width="30.0" height="16.0" rx="3"/><text class="a-t" x="16.0" y="46.0">[Ret]</text><rect class="a-f" x="48.0" y="36.0" width="30.0" height="16.0" rx="3"/><text class="a-t" x="52.0" y="46.0">IsRel</text><rect class="a-g" x="84.0" y="36.0" width="30.0" height="16.0" rx="3"/><text class="a-t" x="88.0" y="46.0">IsSup</text><rect class="a-f" x="120.0" y="36.0" width="30.0" height="16.0" rx="3"/><text class="a-t" x="124.0" y="46.0">IsUse</text><path class="a-b" d="M42.0 44.0L48.0 44.0"/><path class="a-b" d="M78.0 44.0L84.0 44.0"/><path class="a-b" d="M114.0 44.0L120.0 44.0"/></svg></span><span class="al-meta"><span class="al-tag">Retrieval (RAG)</span><span>Advanced</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Self-RAG: the model decides when to retrieve and critiques itself</span><span class="al-d">Self-RAG emits reflection tokens: whether to retrieve, whether each passage is relevant, whether the answer is supported, how useful it is.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/rag-agentic" data-tag="retrieval (rag)" data-q="agentic rag: plan, search in hops, stop when the evidence is enough agentic rag"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><path class="a-b" d="M16.0 70.0 L50.0 30.0 L84.0 58.0 L118.0 22.0 L146.0 40.0"/><circle class="a-db" cx="16.0" cy="70.0" r="2.4"/><circle class="a-db" cx="50.0" cy="30.0" r="2.4"/><circle class="a-db" cx="84.0" cy="58.0" r="2.4"/><circle class="a-db" cx="118.0" cy="22.0" r="2.4"/><circle class="a-db" cx="146.0" cy="40.0" r="2.4"/><circle class="a-d" cx="95.0" cy="7.4" r="0.9"/><circle class="a-d" cx="56.7" cy="46.5" r="0.9"/><circle class="a-d" cx="124.2" cy="45.1" r="0.9"/><circle class="a-d" cx="30.2" cy="51.6" r="0.9"/><circle class="a-d" cx="44.0" cy="70.8" r="0.9"/><circle class="a-d" cx="72.3" cy="82.4" r="0.9"/><circle class="a-d" cx="76.1" cy="23.5" r="0.9"/><circle class="a-d" cx="35.9" cy="37.4" r="0.9"/><circle class="a-d" cx="93.4" cy="70.0" r="0.9"/><circle class="a-d" cx="88.0" cy="29.5" r="0.9"/><circle class="a-d" cx="95.0" cy="14.8" r="0.9"/><circle class="a-d" cx="95.1" cy="29.2" r="0.9"/><circle class="a-d" cx="124.9" cy="65.8" r="0.9"/><circle class="a-d" cx="91.0" cy="23.5" r="0.9"/><circle class="a-d" cx="96.5" cy="52.7" r="0.9"/><circle class="a-d" cx="63.5" cy="31.9" r="0.9"/><circle class="a-d" cx="118.1" cy="36.6" r="0.9"/><circle class="a-d" cx="93.3" cy="74.2" r="0.9"/><circle class="a-d" cx="35.4" cy="82.5" r="0.9"/><circle class="a-d" cx="110.4" cy="31.4" r="0.9"/><circle class="a-d" cx="51.5" cy="19.4" r="0.9"/><circle class="a-d" cx="108.1" cy="-9.0" r="0.9"/><circle class="a-d" cx="108.0" cy="48.5" r="0.9"/><circle class="a-d" cx="59.5" cy="34.3" r="0.9"/><circle class="a-d" cx="69.0" cy="40.2" r="0.9"/><circle class="a-d" cx="106.7" cy="27.5" r="0.9"/><circle class="a-d" cx="112.8" cy="65.7" r="0.9"/><circle class="a-d" cx="72.4" cy="57.8" r="0.9"/><circle class="a-d" cx="63.7" cy="46.2" r="0.9"/><circle class="a-d" cx="95.0" cy="85.7" r="0.9"/><circle class="a-d" cx="78.5" cy="41.8" r="0.9"/><circle class="a-d" cx="81.0" cy="41.2" r="0.9"/><circle class="a-d" cx="47.6" cy="84.5" r="0.9"/><circle class="a-d" cx="61.3" cy="32.7" r="0.9"/><circle class="a-d" cx="58.6" cy="59.5" r="0.9"/><circle class="a-d" cx="169.0" cy="108.8" r="0.9"/><circle class="a-d" cx="109.7" cy="52.3" r="0.9"/><circle class="a-d" cx="79.3" cy="23.9" r="0.9"/><circle class="a-d" cx="102.3" cy="43.1" r="0.9"/><circle class="a-d" cx="142.8" cy="39.2" r="0.9"/><circle class="a-d" cx="49.7" cy="6.3" r="0.9"/><circle class="a-d" cx="26.5" cy="57.9" r="0.9"/><circle class="a-d" cx="90.3" cy="20.7" r="0.9"/><circle class="a-d" cx="33.9" cy="51.0" r="0.9"/><circle class="a-d" cx="69.3" cy="49.8" r="0.9"/><circle class="a-d" cx="79.5" cy="28.5" r="0.9"/><circle class="a-d" cx="27.6" cy="47.8" r="0.9"/><circle class="a-d" cx="83.0" cy="48.9" r="0.9"/><circle class="a-d" cx="108.6" cy="64.6" r="0.9"/><circle class="a-d" cx="65.2" cy="57.4" r="0.9"/></svg></span><span class="al-meta"><span class="al-tag">Retrieval (RAG)</span><span>Advanced</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Agentic RAG: plan, search in hops, stop when the evidence is enough</span><span class="al-d">An agent splits a question into sub-questions, searches hop by hop using what each hop found, and stops when the evidence covers the question.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/graphrag" data-tag="retrieval (rag)" data-q="graphrag: entities, communities and global questions graphrag"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><path class="a-b" d="M40.4 38.7L35.4 37.8"/><path class="a-l" d="M40.4 38.7L49.3 38.3"/><path class="a-l" d="M40.4 38.7L55.6 27.9"/><path class="a-l" d="M40.4 38.7L40.7 29.4"/><path class="a-l" d="M40.4 38.7L32.2 33.5"/><path class="a-l" d="M40.4 38.7L42.2 38.4"/><path class="a-l" d="M40.4 38.7L45.1 52.9"/><path class="a-b" d="M40.4 38.7L48.6 22.2"/><path class="a-l" d="M35.4 37.8L49.3 38.3"/><path class="a-l" d="M35.4 37.8L40.7 29.4"/><path class="a-l" d="M35.4 37.8L32.2 33.5"/><path class="a-l" d="M35.4 37.8L42.2 38.4"/><path class="a-l" d="M35.4 37.8L45.1 52.9"/><path class="a-l" d="M35.4 37.8L48.6 22.2"/><path class="a-b" d="M49.3 38.3L55.6 27.9"/><path class="a-l" d="M49.3 38.3L40.7 29.4"/><path class="a-l" d="M49.3 38.3L32.2 33.5"/><path class="a-l" d="M49.3 38.3L42.2 38.4"/><path class="a-l" d="M49.3 38.3L45.1 52.9"/><path class="a-l" d="M49.3 38.3L48.6 22.2"/><path class="a-l" d="M55.6 27.9L40.7 29.4"/><path class="a-b" d="M55.6 27.9L42.2 38.4"/><path class="a-l" d="M55.6 27.9L48.6 22.2"/><path class="a-l" d="M40.7 29.4L32.2 33.5"/><path class="a-l" d="M40.7 29.4L42.2 38.4"/><path class="a-l" d="M40.7 29.4L48.6 22.2"/><path class="a-l" d="M32.2 33.5L42.2 38.4"/><path class="a-l" d="M32.2 33.5L48.6 22.2"/><path class="a-b" d="M42.2 38.4L45.1 52.9"/><path class="a-l" d="M42.2 38.4L48.6 22.2"/><path class="a-l" d="M112.0 25.0L104.8 40.4"/><path class="a-l" d="M112.0 25.0L107.8 14.4"/><path class="a-l" d="M112.0 25.0L113.1 27.7"/><path class="a-l" d="M112.0 25.0L98.3 22.6"/><path class="a-l" d="M112.0 25.0L103.7 29.8"/><path class="a-b" d="M112.0 25.0L105.7 30.6"/><path class="a-l" d="M112.0 25.0L128.4 23.6"/><path class="a-l" d="M112.0 25.0L101.9 28.0"/><path class="a-l" d="M104.8 40.4L113.1 27.7"/><path class="a-l" d="M104.8 40.4L98.3 22.6"/><path class="a-l" d="M104.8 40.4L103.7 29.8"/><path class="a-l" d="M104.8 40.4L105.7 30.6"/><path class="a-b" d="M104.8 40.4L101.9 28.0"/><path class="a-l" d="M104.8 40.4L96.0 60.4"/><path class="a-l" d="M104.8 40.4L99.3 55.5"/><path class="a-l" d="M107.8 14.4L113.1 27.7"/><path class="a-l" d="M107.8 14.4L98.3 22.6"/><path class="a-l" d="M107.8 14.4L103.7 29.8"/><path class="a-l" d="M107.8 14.4L105.7 30.6"/><path class="a-b" d="M107.8 14.4L101.9 28.0"/><path class="a-l" d="M113.1 27.7L98.3 22.6"/><path class="a-l" d="M113.1 27.7L103.7 29.8"/><path class="a-l" d="M113.1 27.7L105.7 30.6"/><path class="a-l" d="M113.1 27.7L128.4 23.6"/><path class="a-l" d="M113.1 27.7L101.9 28.0"/><path class="a-l" d="M98.3 22.6L103.7 29.8"/><path class="a-b" d="M98.3 22.6L105.7 30.6"/><path class="a-l" d="M98.3 22.6L101.9 28.0"/><path class="a-l" d="M103.7 29.8L105.7 30.6"/><path class="a-l" d="M103.7 29.8L101.9 28.0"/><path class="a-l" d="M105.7 30.6L101.9 28.0"/><path class="a-l" d="M96.0 60.4L99.3 55.5"/><path class="a-l" d="M96.0 60.4L74.7 65.6"/><path class="a-b" d="M96.0 60.4L76.4 61.1"/><path class="a-l" d="M96.0 60.4L89.5 71.8"/><path class="a-l" d="M96.0 60.4L86.1 63.8"/><path class="a-l" d="M96.0 60.4L98.4 69.2"/><path class="a-l" d="M96.0 60.4L82.6 75.7"/><path class="a-l" d="M96.0 60.4L76.0 67.3"/><path class="a-l" d="M99.3 55.5L89.5 71.8"/><path class="a-b" d="M99.3 55.5L86.1 63.8"/><path class="a-l" d="M99.3 55.5L98.4 69.2"/><path class="a-l" d="M74.7 65.6L76.4 61.1"/><path class="a-l" d="M74.7 65.6L89.5 71.8"/><path class="a-l" d="M74.7 65.6L86.1 63.8"/><path class="a-l" d="M74.7 65.6L82.6 75.7"/><path class="a-l" d="M74.7 65.6L76.0 67.3"/><path class="a-b" d="M76.4 61.1L89.5 71.8"/><path class="a-l" d="M76.4 61.1L86.1 63.8"/><path class="a-l" d="M76.4 61.1L82.6 75.7"/><path class="a-l" d="M76.4 61.1L76.0 67.3"/><path class="a-l" d="M89.5 71.8L86.1 63.8"/><path class="a-l" d="M89.5 71.8L98.4 69.2"/><path class="a-l" d="M89.5 71.8L82.6 75.7"/><path class="a-b" d="M89.5 71.8L76.0 67.3"/><path class="a-l" d="M86.1 63.8L98.4 69.2"/><path class="a-l" d="M86.1 63.8L82.6 75.7"/><path class="a-l" d="M86.1 63.8L76.0 67.3"/><path class="a-l" d="M98.4 69.2L82.6 75.7"/><path class="a-l" d="M82.6 75.7L76.0 67.3"/><circle class="a-db" cx="40.0" cy="35.0" r="2.4"/><circle class="a-db" cx="110.0" cy="30.0" r="2.4"/><circle class="a-db" cx="85.0" cy="66.0" r="2.4"/></svg></span><span class="al-meta"><span class="al-tag">Retrieval (RAG)</span><span>Advanced</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">GraphRAG: entities, communities and global questions</span><span class="al-d">Build a knowledge graph from the chunks, group entities into communities with summaries, and answer questions about the whole corpus or across documents.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/chunking-strategies" data-tag="retrieval (rag)" data-q="chunking strategies: size, overlap and boundaries chunking"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"></svg></span><span class="al-meta"><span class="al-tag">Retrieval (RAG)</span><span>Beginner</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Chunking strategies: size, overlap and boundaries</span><span class="al-d">Split a document with fixed-size, sentence or recursive chunking, change size and overlap, and see whether the answer survives in one chunk and ranks first.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/embedding-space" data-tag="retrieval (rag)" data-q="embedding space: where your query lands among the documents embeddings"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><circle class="a-d" cx="31.4" cy="31.9" r="0.9"/><circle class="a-d" cx="47.0" cy="27.5" r="0.9"/><circle class="a-d" cx="30.7" cy="29.7" r="0.9"/><circle class="a-d" cx="43.3" cy="35.4" r="0.9"/><circle class="a-d" cx="31.5" cy="23.7" r="0.9"/><circle class="a-d" cx="44.9" cy="32.3" r="0.9"/><circle class="a-d" cx="50.7" cy="34.8" r="0.9"/><circle class="a-d" cx="39.3" cy="28.7" r="0.9"/><circle class="a-d" cx="58.0" cy="28.6" r="0.9"/><circle class="a-d" cx="34.5" cy="24.4" r="0.9"/><circle class="a-d" cx="40.7" cy="30.7" r="0.9"/><circle class="a-d" cx="37.6" cy="29.7" r="0.9"/><circle class="a-d" cx="41.9" cy="33.4" r="0.9"/><circle class="a-d" cx="47.7" cy="31.0" r="0.9"/><circle class="a-d" cx="27.5" cy="33.4" r="0.9"/><circle class="a-d" cx="30.1" cy="29.2" r="0.9"/><circle class="a-d" cx="29.7" cy="30.0" r="0.9"/><circle class="a-d" cx="34.6" cy="31.0" r="0.9"/><circle class="a-d" cx="63.1" cy="29.8" r="0.9"/><circle class="a-d" cx="45.9" cy="23.5" r="0.9"/><circle class="a-d" cx="38.0" cy="33.2" r="0.9"/><circle class="a-d" cx="39.4" cy="31.8" r="0.9"/><circle class="a-d" cx="104.9" cy="29.1" r="0.9"/><circle class="a-d" cx="113.1" cy="25.6" r="0.9"/><circle class="a-d" cx="121.8" cy="28.3" r="0.9"/><circle class="a-d" cx="113.1" cy="36.8" r="0.9"/><circle class="a-d" cx="111.5" cy="21.8" r="0.9"/><circle class="a-d" cx="105.4" cy="25.4" r="0.9"/><circle class="a-d" cx="119.6" cy="32.2" r="0.9"/><circle class="a-d" cx="123.0" cy="37.1" r="0.9"/><circle class="a-d" cx="127.1" cy="31.3" r="0.9"/><circle class="a-d" cx="124.5" cy="31.4" r="0.9"/><circle class="a-d" cx="111.3" cy="32.3" r="0.9"/><circle class="a-d" cx="115.7" cy="27.2" r="0.9"/><circle class="a-d" cx="129.4" cy="32.1" r="0.9"/><circle class="a-d" cx="111.9" cy="27.9" r="0.9"/><circle class="a-d" cx="131.9" cy="27.0" r="0.9"/><circle class="a-d" cx="120.6" cy="31.3" r="0.9"/><circle class="a-d" cx="119.0" cy="31.2" r="0.9"/><circle class="a-d" cx="112.0" cy="26.0" r="0.9"/><circle class="a-d" cx="114.3" cy="26.5" r="0.9"/><circle class="a-d" cx="121.1" cy="35.9" r="0.9"/><circle class="a-d" cx="109.1" cy="31.0" r="0.9"/><circle class="a-d" cx="108.4" cy="24.8" r="0.9"/><circle class="a-d" cx="69.4" cy="72.4" r="0.9"/><circle class="a-d" cx="64.5" cy="56.3" r="0.9"/><circle class="a-d" cx="72.0" cy="68.3" r="0.9"/><circle class="a-d" cx="73.0" cy="67.8" r="0.9"/><circle class="a-d" cx="69.6" cy="62.4" r="0.9"/><circle class="a-d" cx="65.2" cy="64.3" r="0.9"/><circle class="a-d" cx="67.9" cy="70.0" r="0.9"/><circle class="a-d" cx="75.0" cy="59.3" r="0.9"/><circle class="a-d" cx="75.6" cy="65.4" r="0.9"/><circle class="a-d" cx="78.6" cy="72.9" r="0.9"/><circle class="a-d" cx="74.6" cy="56.7" r="0.9"/><circle class="a-d" cx="75.3" cy="57.8" r="0.9"/><circle class="a-d" cx="81.8" cy="62.3" r="0.9"/><circle class="a-d" cx="69.4" cy="74.5" r="0.9"/><circle class="a-d" cx="58.3" cy="65.7" r="0.9"/><circle class="a-d" cx="63.2" cy="63.0" r="0.9"/><circle class="a-d" cx="65.1" cy="68.9" r="0.9"/><circle class="a-d" cx="68.6" cy="67.3" r="0.9"/><circle class="a-d" cx="75.3" cy="57.1" r="0.9"/><circle class="a-d" cx="80.7" cy="64.3" r="0.9"/><circle class="a-d" cx="59.0" cy="62.3" r="0.9"/><circle class="a-d" cx="77.0" cy="70.5" r="0.9"/><circle class="a-d" cx="126.1" cy="69.4" r="0.9"/><circle class="a-d" cx="113.1" cy="59.2" r="0.9"/><circle class="a-d" cx="119.5" cy="68.7" r="0.9"/><circle class="a-d" cx="119.3" cy="60.9" r="0.9"/><circle class="a-d" cx="117.5" cy="71.0" r="0.9"/><circle class="a-d" cx="113.1" cy="69.7" r="0.9"/><circle class="a-d" cx="122.6" cy="67.7" r="0.9"/><circle class="a-d" cx="123.8" cy="64.1" r="0.9"/><circle class="a-d" cx="119.4" cy="72.3" r="0.9"/><circle class="a-d" cx="118.1" cy="64.5" r="0.9"/><circle class="a-d" cx="113.0" cy="64.4" r="0.9"/><circle class="a-d" cx="123.7" cy="70.4" r="0.9"/><circle class="a-d" cx="123.2" cy="68.2" r="0.9"/><circle class="a-d" cx="122.9" cy="67.2" r="0.9"/><circle class="a-db" cx="105.0" cy="34.0" r="3.0"/><path class="a-b" d="M105.0 34.0L112.0 26.0"/><path class="a-b" d="M105.0 34.0L118.0 31.0"/></svg></span><span class="al-meta"><span class="al-tag">Retrieval (RAG)</span><span>Beginner</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Embedding space: where your query lands among the documents</span><span class="al-d">Every chunk becomes a point in a vector space. Type a query, see where it lands on a 2D map and which chunks are its nearest neighbours by cosine.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/hnsw-vector-search" data-tag="retrieval (rag)" data-q="hnsw: how a vector database finds neighbours fast hnsw search"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><polygon class="a-g" points="30,8 140,8 130,24 20,24"/><polygon class="a-g" points="30,34 140,34 130,50 20,50"/><polygon class="a-g" points="30,60 140,60 130,76 20,76"/><path class="a-b" d="M40.0 14.0 L110.0 18.0 L100.0 42.0 L70.0 46.0 L78.0 70.0 L92.0 68.0"/><circle class="a-db" cx="40.0" cy="14.0" r="1.6"/><circle class="a-db" cx="110.0" cy="18.0" r="1.6"/><circle class="a-db" cx="100.0" cy="42.0" r="1.6"/><circle class="a-db" cx="70.0" cy="46.0" r="1.6"/><circle class="a-db" cx="78.0" cy="70.0" r="1.6"/><circle class="a-db" cx="92.0" cy="68.0" r="1.6"/><circle class="a-d" cx="59.3" cy="71.4" r="0.9"/><circle class="a-d" cx="108.5" cy="91.6" r="0.9"/><circle class="a-d" cx="106.0" cy="68.7" r="0.9"/><circle class="a-d" cx="87.2" cy="64.2" r="0.9"/><circle class="a-d" cx="51.3" cy="60.9" r="0.9"/><circle class="a-d" cx="100.8" cy="71.8" r="0.9"/><circle class="a-d" cx="73.0" cy="51.8" r="0.9"/><circle class="a-d" cx="78.0" cy="59.1" r="0.9"/><circle class="a-d" cx="80.8" cy="75.4" r="0.9"/><circle class="a-d" cx="30.1" cy="66.6" r="0.9"/><circle class="a-d" cx="48.0" cy="53.7" r="0.9"/><circle class="a-d" cx="52.9" cy="85.0" r="0.9"/><circle class="a-d" cx="94.7" cy="75.6" r="0.9"/><circle class="a-d" cx="63.8" cy="46.0" r="0.9"/><circle class="a-d" cx="71.7" cy="71.1" r="0.9"/><circle class="a-d" cx="78.4" cy="78.5" r="0.9"/><circle class="a-d" cx="72.2" cy="95.5" r="0.9"/><circle class="a-d" cx="94.3" cy="98.5" r="0.9"/><circle class="a-d" cx="85.4" cy="64.4" r="0.9"/><circle class="a-d" cx="61.5" cy="80.4" r="0.9"/><circle class="a-d" cx="102.9" cy="70.4" r="0.9"/><circle class="a-d" cx="88.9" cy="63.8" r="0.9"/><circle class="a-d" cx="70.8" cy="63.5" r="0.9"/><circle class="a-d" cx="38.9" cy="52.5" r="0.9"/><circle class="a-d" cx="80.8" cy="69.9" r="0.9"/><circle class="a-d" cx="103.7" cy="77.6" r="0.9"/><circle class="a-d" cx="89.2" cy="68.7" r="0.9"/><circle class="a-d" cx="30.5" cy="68.7" r="0.9"/><circle class="a-d" cx="60.6" cy="75.5" r="0.9"/><circle class="a-d" cx="73.7" cy="62.8" r="0.9"/></svg></span><span class="al-meta"><span class="al-tag">Retrieval (RAG)</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">HNSW: how a vector database finds neighbours fast</span><span class="al-d">Watch a Hierarchical Navigable Small World index answer a nearest-neighbour query in a few hops instead of comparing against every point.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/rag-evaluation-ragas" data-tag="retrieval (rag)" data-q="evaluating rag: faithfulness, context precision and recall rag evaluation"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-g" x="20.0" y="14.0" width="120.0" height="5.0" rx="2.5"/><rect class="a-f" x="20.0" y="14.0" width="104.0" height="5.0" rx="2.5"/><text class="a-t" x="20.0" y="12.0">faithfulness</text><rect class="a-g" x="20.0" y="31.0" width="120.0" height="5.0" rx="2.5"/><rect class="a-f" x="20.0" y="31.0" width="82.0" height="5.0" rx="2.5"/><text class="a-t" x="20.0" y="29.0">context precision</text><rect class="a-g" x="20.0" y="48.0" width="120.0" height="5.0" rx="2.5"/><rect class="a-f" x="20.0" y="48.0" width="96.0" height="5.0" rx="2.5"/><text class="a-t" x="20.0" y="46.0">context recall</text><rect class="a-g" x="20.0" y="65.0" width="120.0" height="5.0" rx="2.5"/><rect class="a-f" x="20.0" y="65.0" width="110.0" height="5.0" rx="2.5"/><text class="a-t" x="20.0" y="63.0">answer relevancy</text></svg></span><span class="al-meta"><span class="al-tag">Retrieval (RAG)</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-lab">Code lab</span></span><span class="al-t">Evaluating RAG: faithfulness, context precision and recall</span><span class="al-d">Score RAG answers the RAGAS way: break the answer into claims, check each against the contexts, and measure context precision, recall and answer relevancy.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a></div></section><section class="al-month"><h2 class="al-mh" id="agents">Agents <span>6</span></h2><div class="al-grid"><a class="al-card" href="/ai-simulations/agent-run" data-tag="agents" data-q="an agent run: langgraph loop, langfuse trace, deepeval and ragas scores agent run"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-g" x="12.0" y="38.0" width="24.0" height="14.0" rx="4"/><rect class="a-g" x="58.0" y="36.0" width="34.0" height="14.0" rx="4"/><rect class="a-g" x="114.0" y="14.0" width="30.0" height="14.0" rx="4"/><rect class="a-g" x="114.0" y="60.0" width="30.0" height="14.0" rx="4"/><path class="a-l" d="M36.0 45.0L58.0 43.0"/><path class="a-b" d="M92.0 40.0 L114.0 21.0"/><path class="a-l" d="M114.0 26.0 L94.0 46.0"/><path class="a-l" d="M92.0 48.0L114.0 67.0"/><circle class="a-db" cx="103.0" cy="30.0" r="1.8"/></svg></span><span class="al-meta"><span class="al-tag">Agents</span><span>Intermediate</span><span class="al-script">Scripted run</span><span class="al-lab">Code lab</span></span><span class="al-t">An agent run: LangGraph loop, Langfuse trace, DeepEval and RAGAS scores</span><span class="al-d">Follow an AI agent step by step: the LangGraph loop between model and tools, the trace as spans with time and tokens, and evaluation scores.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/agentic-patterns" data-tag="agents" data-q="agentic patterns: react, plan and execute, reflection, multi-agent, routing agentic patterns"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-f" x="64.0" y="38.0" width="32.0" height="14.0" rx="4"/><rect class="a-g" x="14.0" y="12.0" width="26.0" height="11.0" rx="3"/><path class="a-l" d="M80.0 45.0L27.0 17.5"/><rect class="a-g" x="14.0" y="66.0" width="26.0" height="11.0" rx="3"/><path class="a-l" d="M80.0 45.0L27.0 71.5"/><rect class="a-g" x="122.0" y="12.0" width="26.0" height="11.0" rx="3"/><path class="a-l" d="M80.0 45.0L135.0 17.5"/><rect class="a-g" x="122.0" y="66.0" width="26.0" height="11.0" rx="3"/><path class="a-l" d="M80.0 45.0L135.0 71.5"/><rect class="a-g" x="122.0" y="39.0" width="26.0" height="11.0" rx="3"/><path class="a-l" d="M80.0 45.0L135.0 44.5"/><rect class="a-g" x="14.0" y="39.0" width="26.0" height="11.0" rx="3"/><path class="a-l" d="M80.0 45.0L27.0 44.5"/><path class="a-b" d="M80.0 45.0L135.0 17.0"/></svg></span><span class="al-meta"><span class="al-tag">Agents</span><span>Beginner</span><span class="al-script">Scripted run</span><span class="al-lab">Code lab</span></span><span class="al-t">Agentic patterns: ReAct, plan and execute, reflection, multi-agent, routing</span><span class="al-d">Six ways to wire models, tools and people: ReAct, plan and execute, reflection, orchestrator and workers, routing and human approval.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/langgraph-state-checkpoints" data-tag="agents" data-q="langgraph state, checkpoints, interrupts and time travel langgraph state"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-g" x="8.0" y="30.0" width="24.0" height="14.0" rx="3"/><path class="a-l" d="M32.0 37.0L38.0 37.0"/><rect class="a-g" x="38.0" y="30.0" width="24.0" height="14.0" rx="3"/><path class="a-l" d="M62.0 37.0L68.0 37.0"/><rect class="a-g" x="68.0" y="30.0" width="24.0" height="14.0" rx="3"/><path class="a-l" d="M92.0 37.0L98.0 37.0"/><rect class="a-f" x="98.0" y="30.0" width="24.0" height="14.0" rx="3"/><path class="a-l" d="M122.0 37.0L128.0 37.0"/><rect class="a-g" x="128.0" y="30.0" width="24.0" height="14.0" rx="3"/><circle class="a-db" cx="20.0" cy="66.0" r="2.0"/><circle class="a-db" cx="50.0" cy="66.0" r="2.0"/><circle class="a-db" cx="80.0" cy="66.0" r="2.0"/><circle class="a-db" cx="110.0" cy="66.0" r="2.0"/><circle class="a-db" cx="140.0" cy="66.0" r="2.0"/><path class="a-l2" d="M14.0 66.0L146.0 66.0"/><path class="a-b" d="M80.0 66.0 L80.0 76.0 L140.0 76.0"/></svg></span><span class="al-meta"><span class="al-tag">Agents</span><span>Intermediate</span><span class="al-script">Scripted run</span><span class="al-lab">Code lab</span></span><span class="al-t">LangGraph state, checkpoints, interrupts and time travel</span><span class="al-d">Step through a LangGraph run: the state after every node, saved checkpoints, an interrupt for human approval, and time travel from an earlier checkpoint.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/react-tool-calling" data-tag="agents" data-q="react agent: thought, tool call, observation, with your own question react tool calling"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-g" x="14.0" y="36.0" width="26.0" height="18.0" rx="4"/><path class="a-b" d="M40.0 45.0L48.0 45.0"/><rect class="a-f" x="48.0" y="36.0" width="26.0" height="18.0" rx="4"/><path class="a-b" d="M74.0 45.0L82.0 45.0"/><rect class="a-g" x="82.0" y="36.0" width="26.0" height="18.0" rx="4"/><path class="a-b" d="M108.0 45.0L116.0 45.0"/><rect class="a-f" x="116.0" y="36.0" width="26.0" height="18.0" rx="4"/><text class="a-t" x="17.0" y="47.0">think</text><text class="a-t" x="50.0" y="47.0">act</text><text class="a-t" x="84.0" y="47.0">obs</text><text class="a-t" x="116.0" y="47.0">think</text><path class="a-l" d="M127.0 54.0 L127.0 70.0 L27.0 70.0 L27.0 54.0"/></svg></span><span class="al-meta"><span class="al-tag">Agents</span><span>Intermediate</span><span class="al-real">Real tools, rule-based planner</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">ReAct agent: thought, tool call, observation, with your own question</span><span class="al-d">Ask your own question and watch an agent plan, call a calculator, unit converter, date tool or knowledge-base search, read the result and answer.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/multi-agent-supervisor" data-tag="agents" data-q="multi-agent supervisor: researcher, analyst, writer and reviewer multi-agent team"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-f" x="62.0" y="10.0" width="36.0" height="14.0" rx="4"/><rect class="a-g" x="14.0" y="62.0" width="30.0" height="14.0" rx="4"/><path class="a-l" d="M80.0 24.0L29.0 62.0"/><rect class="a-g" x="48.0" y="62.0" width="30.0" height="14.0" rx="4"/><path class="a-l" d="M80.0 24.0L63.0 62.0"/><rect class="a-g" x="82.0" y="62.0" width="30.0" height="14.0" rx="4"/><path class="a-l" d="M80.0 24.0L97.0 62.0"/><rect class="a-g" x="116.0" y="62.0" width="30.0" height="14.0" rx="4"/><path class="a-l" d="M80.0 24.0L131.0 62.0"/></svg></span><span class="al-meta"><span class="al-tag">Agents</span><span>Advanced</span><span class="al-real">Real tools, rule-based planner</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Multi-agent supervisor: researcher, analyst, writer and reviewer</span><span class="al-d">Give a team of agents your own task: a supervisor routes it to a researcher, an analyst and a writer, and a reviewer sends unsupported drafts back.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/guardrails" data-tag="agents" data-q="guardrails: prompt injection, pii masking and grounding checks guardrails"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-g" x="10.0" y="32.0" width="28.0" height="26.0" rx="4"/><rect class="a-f" x="66.0" y="30.0" width="28.0" height="30.0" rx="4"/><rect class="a-g" x="122.0" y="32.0" width="28.0" height="26.0" rx="4"/><path class="a-b" d="M38.0 45.0L52.0 45.0"/><path class="a-b" d="M94.0 45.0L108.0 45.0"/><path class="a-l" d="M52.0 18.0L52.0 72.0"/><path class="a-l" d="M56.0 18.0L56.0 72.0"/><path class="a-l" d="M108.0 18.0L108.0 72.0"/><path class="a-l" d="M112.0 18.0L112.0 72.0"/><path class="a-l2" d="M38.0 50.0L52.0 66.0"/></svg></span><span class="al-meta"><span class="al-tag">Agents</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Guardrails: prompt injection, PII masking and grounding checks</span><span class="al-d">Type any message and watch input and output guardrails detect prompt injection, mask cards and emails, stop off-topic requests and remove unsupported sentences.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a></div></section><section class="al-month"><h2 class="al-mh" id="large-tabular-foundation-model">Large tabular foundation model <span>11</span></h2><div class="al-grid"><a class="al-card" href="/ai-simulations/tfm-classification" data-tag="large tabular foundation model" data-q="in-context classification: a tabular foundation model reads labelled rows in-context classification"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-f" x="12.0" y="12.0" width="60.0" height="6.0" rx="1"/><rect class="a-g" x="12.0" y="20.0" width="60.0" height="6.0" rx="1"/><rect class="a-g" x="12.0" y="28.0" width="60.0" height="6.0" rx="1"/><rect class="a-f" x="12.0" y="36.0" width="60.0" height="6.0" rx="1"/><rect class="a-g" x="12.0" y="44.0" width="60.0" height="6.0" rx="1"/><rect class="a-g" x="12.0" y="52.0" width="60.0" height="6.0" rx="1"/><rect class="a-f" x="12.0" y="60.0" width="60.0" height="6.0" rx="1"/><rect class="a-g" x="12.0" y="68.0" width="60.0" height="6.0" rx="1"/><rect class="a-g" x="12.0" y="76.0" width="60.0" height="6.0" rx="1"/><path class="a-b" d="M74.0 15.0L120.0 46.0"/><path class="a-l2" d="M74.0 23.0L120.0 46.0"/><path class="a-l2" d="M74.0 31.0L120.0 46.0"/><path class="a-b" d="M74.0 39.0L120.0 46.0"/><path class="a-l2" d="M74.0 47.0L120.0 46.0"/><path class="a-l2" d="M74.0 55.0L120.0 46.0"/><path class="a-b" d="M74.0 63.0L120.0 46.0"/><path class="a-l2" d="M74.0 71.0L120.0 46.0"/><path class="a-l2" d="M74.0 79.0L120.0 46.0"/><circle class="a-db" cx="124.0" cy="46.0" r="3.0"/><rect class="a-f" x="132.0" y="38.0" width="18.0" height="16.0" rx="3"/></svg></span><span class="al-meta"><span class="al-tag">Large tabular foundation model</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">In-context classification: a tabular foundation model reads labelled rows</span><span class="al-d">Watch a tabular foundation model classify rows with no training: query rows attend to labelled rows, with attention, feature relevance and a learning curve.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/tfm-regression" data-tag="large tabular foundation model" data-q="in-context regression: predicting numbers from the rows of the table in-context regression"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><path class="a-l2" d="M10.0 20.0L150.0 20.0"/><path class="a-l2" d="M10.0 38.0L150.0 38.0"/><path class="a-l2" d="M10.0 56.0L150.0 56.0"/><path class="a-l2" d="M10.0 74.0L150.0 74.0"/><path class="a-f" d="M12 70 C40 60 60 40 80 34 S120 22 148 18 L148 34 C120 38 100 50 80 52 S40 76 12 82 Z"/><path class="a-b" d="M12.0 76.0 L13.4 74.3 L14.7 72.6 L16.1 70.9 L17.4 69.3 L18.8 67.8 L20.2 66.3 L21.5 64.8 L22.9 63.4 L24.2 62.0 L25.6 60.7 L27.0 59.5 L28.3 58.3 L29.7 57.1 L31.0 56.0 L32.4 55.0 L33.8 54.0 L35.1 53.1 L36.5 52.2 L37.8 51.4 L39.2 50.6 L40.6 49.9 L41.9 49.2 L43.3 48.6 L44.6 48.0 L46.0 47.5 L47.4 47.1 L48.7 46.6 L50.1 46.3 L51.4 46.0 L52.8 45.7 L54.2 45.4 L55.5 45.2 L56.9 45.1 L58.2 45.0 L59.6 44.9 L61.0 44.8 L62.3 44.8 L63.7 44.8 L65.0 44.8 L66.4 44.9 L67.8 44.9 L69.1 45.0 L70.5 45.1 L71.8 45.2 L73.2 45.4 L74.6 45.5 L75.9 45.6 L77.3 45.8 L78.6 45.9 L80.0 46.1 L81.4 46.2 L82.7 46.3 L84.1 46.4 L85.4 46.5 L86.8 46.6 L88.2 46.7 L89.5 46.8 L90.9 46.8 L92.2 46.9 L93.6 46.9 L95.0 46.9 L96.3 46.8 L97.7 46.8 L99.0 46.7 L100.4 46.6 L101.8 46.4 L103.1 46.3 L104.5 46.1 L105.8 45.9 L107.2 45.6 L108.6 45.4 L109.9 45.1 L111.3 44.8 L112.6 44.4 L114.0 44.1 L115.4 43.7 L116.7 43.3 L118.1 42.9 L119.4 42.5 L120.8 42.0 L122.2 41.6 L123.5 41.1 L124.9 40.6 L126.2 40.1 L127.6 39.6 L129.0 39.1 L130.3 38.6 L131.7 38.1 L133.0 37.6 L134.4 37.1 L135.8 36.6 L137.1 36.2 L138.5 35.7 L139.8 35.2 L141.2 34.8 L142.6 34.4 L143.9 34.0 L145.3 33.6 L146.6 33.2 L148.0 32.9"/><circle class="a-d" cx="81.1" cy="64.2" r="0.9"/><circle class="a-d" cx="101.5" cy="70.3" r="0.9"/><circle class="a-d" cx="83.3" cy="65.4" r="0.9"/><circle class="a-d" cx="99.8" cy="39.1" r="0.9"/><circle class="a-d" cx="115.0" cy="41.0" r="0.9"/><circle class="a-d" cx="50.4" cy="39.5" r="0.9"/><circle class="a-d" cx="62.2" cy="38.5" r="0.9"/><circle class="a-d" cx="102.4" cy="45.7" r="0.9"/><circle class="a-d" cx="122.8" cy="42.9" r="0.9"/><circle class="a-d" cx="96.7" cy="48.1" r="0.9"/><circle class="a-d" cx="40.0" cy="49.9" r="0.9"/><circle class="a-d" cx="93.4" cy="65.1" r="0.9"/><circle class="a-d" cx="54.1" cy="66.0" r="0.9"/><circle class="a-d" cx="113.4" cy="34.4" r="0.9"/><circle class="a-d" cx="93.0" cy="70.4" r="0.9"/><circle class="a-d" cx="41.2" cy="57.1" r="0.9"/><circle class="a-d" cx="32.9" cy="52.3" r="0.9"/><circle class="a-d" cx="120.9" cy="43.5" r="0.9"/><circle class="a-d" cx="98.7" cy="20.7" r="0.9"/><circle class="a-d" cx="68.6" cy="40.8" r="0.9"/><circle class="a-d" cx="104.5" cy="44.5" r="0.9"/><circle class="a-d" cx="103.4" cy="47.7" r="0.9"/><circle class="a-d" cx="67.3" cy="51.3" r="0.9"/><circle class="a-d" cx="90.4" cy="44.1" r="0.9"/><circle class="a-d" cx="102.9" cy="35.0" r="0.9"/><circle class="a-d" cx="89.0" cy="59.1" r="0.9"/></svg></span><span class="al-meta"><span class="al-tag">Large tabular foundation model</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">In-context regression: predicting numbers from the rows of the table</span><span class="al-d">Predict house prices with in-context learning: the query attends to similar rows and reads off a value and an interval, compared with ridge regression and kNN.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/tfm-probabilistic-regression" data-tag="large tabular foundation model" data-q="probabilistic regression: the whole predictive distribution, scored properly probabilistic regression"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><path class="a-l2" d="M12.0 42.0 L13.4 41.7 L14.7 41.4 L16.1 41.1 L17.4 40.8 L18.8 40.5 L20.2 40.2 L21.5 39.9 L22.9 39.6 L24.2 39.3 L25.6 39.0 L27.0 38.7 L28.3 38.4 L29.7 38.1 L31.0 37.8 L32.4 37.5 L33.8 37.2 L35.1 36.9 L36.5 36.6 L37.8 36.3 L39.2 36.0 L40.6 35.7 L41.9 35.4 L43.3 35.1 L44.6 34.8 L46.0 34.5 L47.4 34.2 L48.7 33.9 L50.1 33.6 L51.4 33.3 L52.8 33.0 L54.2 32.7 L55.5 32.4 L56.9 32.1 L58.2 31.8 L59.6 31.5 L61.0 31.2 L62.3 30.9 L63.7 30.6 L65.0 30.3 L66.4 30.0 L67.8 29.7 L69.1 29.4 L70.5 29.1 L71.8 28.8 L73.2 28.5 L74.6 28.2 L75.9 27.9 L77.3 27.6 L78.6 27.3 L80.0 27.0 L81.4 26.7 L82.7 26.4 L84.1 26.1 L85.4 25.8 L86.8 25.5 L88.2 25.2 L89.5 24.9 L90.9 24.6 L92.2 24.3 L93.6 24.0 L95.0 23.7 L96.3 23.4 L97.7 23.1 L99.0 22.8 L100.4 22.5 L101.8 22.2 L103.1 21.9 L104.5 21.6 L105.8 21.3 L107.2 21.0 L108.6 20.7 L109.9 20.4 L111.3 20.1 L112.6 19.8 L114.0 19.5 L115.4 19.2 L116.7 18.9 L118.1 18.6 L119.4 18.3 L120.8 18.0 L122.2 17.7 L123.5 17.4 L124.9 17.1 L126.2 16.8 L127.6 16.5 L129.0 16.2 L130.3 15.9 L131.7 15.6 L133.0 15.3 L134.4 15.0 L135.8 14.7 L137.1 14.4 L138.5 14.1 L139.8 13.8 L141.2 13.5 L142.6 13.2 L143.9 12.9 L145.3 12.6 L146.6 12.3 L148.0 12.0"/><path class="a-l" d="M12.0 44.0 L13.4 43.9 L14.7 43.7 L16.1 43.5 L17.4 43.4 L18.8 43.2 L20.2 43.1 L21.5 43.0 L22.9 42.8 L24.2 42.6 L25.6 42.5 L27.0 42.4 L28.3 42.2 L29.7 42.0 L31.0 41.9 L32.4 41.8 L33.8 41.6 L35.1 41.5 L36.5 41.3 L37.8 41.1 L39.2 41.0 L40.6 40.9 L41.9 40.7 L43.3 40.5 L44.6 40.4 L46.0 40.2 L47.4 40.1 L48.7 40.0 L50.1 39.8 L51.4 39.6 L52.8 39.5 L54.2 39.4 L55.5 39.2 L56.9 39.0 L58.2 38.9 L59.6 38.8 L61.0 38.6 L62.3 38.5 L63.7 38.3 L65.0 38.1 L66.4 38.0 L67.8 37.9 L69.1 37.7 L70.5 37.5 L71.8 37.4 L73.2 37.2 L74.6 37.1 L75.9 37.0 L77.3 36.8 L78.6 36.6 L80.0 36.5 L81.4 36.4 L82.7 36.2 L84.1 36.0 L85.4 35.9 L86.8 35.8 L88.2 35.6 L89.5 35.5 L90.9 35.3 L92.2 35.1 L93.6 35.0 L95.0 34.9 L96.3 34.7 L97.7 34.5 L99.0 34.4 L100.4 34.2 L101.8 34.1 L103.1 34.0 L104.5 33.8 L105.8 33.6 L107.2 33.5 L108.6 33.4 L109.9 33.2 L111.3 33.0 L112.6 32.9 L114.0 32.8 L115.4 32.6 L116.7 32.5 L118.1 32.3 L119.4 32.1 L120.8 32.0 L122.2 31.9 L123.5 31.7 L124.9 31.6 L126.2 31.4 L127.6 31.2 L129.0 31.1 L130.3 31.0 L131.7 30.8 L133.0 30.6 L134.4 30.5 L135.8 30.4 L137.1 30.2 L138.5 30.1 L139.8 29.9 L141.2 29.8 L142.6 29.6 L143.9 29.5 L145.3 29.3 L146.6 29.1 L148.0 29.0"/><path class="a-l" d="M12.0 48.0 L13.4 48.1 L14.7 48.3 L16.1 48.5 L17.4 48.6 L18.8 48.8 L20.2 48.9 L21.5 49.0 L22.9 49.2 L24.2 49.4 L25.6 49.5 L27.0 49.6 L28.3 49.8 L29.7 50.0 L31.0 50.1 L32.4 50.2 L33.8 50.4 L35.1 50.5 L36.5 50.7 L37.8 50.9 L39.2 51.0 L40.6 51.1 L41.9 51.3 L43.3 51.5 L44.6 51.6 L46.0 51.8 L47.4 51.9 L48.7 52.0 L50.1 52.2 L51.4 52.4 L52.8 52.5 L54.2 52.6 L55.5 52.8 L56.9 53.0 L58.2 53.1 L59.6 53.2 L61.0 53.4 L62.3 53.5 L63.7 53.7 L65.0 53.9 L66.4 54.0 L67.8 54.1 L69.1 54.3 L70.5 54.5 L71.8 54.6 L73.2 54.8 L74.6 54.9 L75.9 55.0 L77.3 55.2 L78.6 55.4 L80.0 55.5 L81.4 55.6 L82.7 55.8 L84.1 56.0 L85.4 56.1 L86.8 56.2 L88.2 56.4 L89.5 56.5 L90.9 56.7 L92.2 56.9 L93.6 57.0 L95.0 57.1 L96.3 57.3 L97.7 57.5 L99.0 57.6 L100.4 57.8 L101.8 57.9 L103.1 58.0 L104.5 58.2 L105.8 58.4 L107.2 58.5 L108.6 58.6 L109.9 58.8 L111.3 59.0 L112.6 59.1 L114.0 59.2 L115.4 59.4 L116.7 59.5 L118.1 59.7 L119.4 59.9 L120.8 60.0 L122.2 60.1 L123.5 60.3 L124.9 60.5 L126.2 60.6 L127.6 60.8 L129.0 60.9 L130.3 61.0 L131.7 61.2 L133.0 61.4 L134.4 61.5 L135.8 61.6 L137.1 61.8 L138.5 62.0 L139.8 62.1 L141.2 62.2 L142.6 62.4 L143.9 62.5 L145.3 62.7 L146.6 62.9 L148.0 63.0"/><path class="a-l2" d="M12.0 50.0 L13.4 50.3 L14.7 50.6 L16.1 50.9 L17.4 51.2 L18.8 51.5 L20.2 51.8 L21.5 52.1 L22.9 52.4 L24.2 52.7 L25.6 53.0 L27.0 53.3 L28.3 53.6 L29.7 53.9 L31.0 54.2 L32.4 54.5 L33.8 54.8 L35.1 55.1 L36.5 55.4 L37.8 55.7 L39.2 56.0 L40.6 56.3 L41.9 56.6 L43.3 56.9 L44.6 57.2 L46.0 57.5 L47.4 57.8 L48.7 58.1 L50.1 58.4 L51.4 58.7 L52.8 59.0 L54.2 59.3 L55.5 59.6 L56.9 59.9 L58.2 60.2 L59.6 60.5 L61.0 60.8 L62.3 61.1 L63.7 61.4 L65.0 61.7 L66.4 62.0 L67.8 62.3 L69.1 62.6 L70.5 62.9 L71.8 63.2 L73.2 63.5 L74.6 63.8 L75.9 64.1 L77.3 64.4 L78.6 64.7 L80.0 65.0 L81.4 65.3 L82.7 65.6 L84.1 65.9 L85.4 66.2 L86.8 66.5 L88.2 66.8 L89.5 67.1 L90.9 67.4 L92.2 67.7 L93.6 68.0 L95.0 68.3 L96.3 68.6 L97.7 68.9 L99.0 69.2 L100.4 69.5 L101.8 69.8 L103.1 70.1 L104.5 70.4 L105.8 70.7 L107.2 71.0 L108.6 71.3 L109.9 71.6 L111.3 71.9 L112.6 72.2 L114.0 72.5 L115.4 72.8 L116.7 73.1 L118.1 73.4 L119.4 73.7 L120.8 74.0 L122.2 74.3 L123.5 74.6 L124.9 74.9 L126.2 75.2 L127.6 75.5 L129.0 75.8 L130.3 76.1 L131.7 76.4 L133.0 76.7 L134.4 77.0 L135.8 77.3 L137.1 77.6 L138.5 77.9 L139.8 78.2 L141.2 78.5 L142.6 78.8 L143.9 79.1 L145.3 79.4 L146.6 79.7 L148.0 80.0"/><path class="a-b" d="M12.0 46.0 L13.4 46.0 L14.7 46.0 L16.1 46.0 L17.4 46.0 L18.8 46.0 L20.2 46.0 L21.5 46.0 L22.9 46.0 L24.2 46.0 L25.6 46.0 L27.0 46.0 L28.3 46.0 L29.7 46.0 L31.0 46.0 L32.4 46.0 L33.8 46.0 L35.1 46.0 L36.5 46.0 L37.8 46.0 L39.2 46.0 L40.6 46.0 L41.9 46.0 L43.3 46.0 L44.6 46.0 L46.0 46.0 L47.4 46.0 L48.7 46.0 L50.1 46.0 L51.4 46.0 L52.8 46.0 L54.2 46.0 L55.5 46.0 L56.9 46.0 L58.2 46.0 L59.6 46.0 L61.0 46.0 L62.3 46.0 L63.7 46.0 L65.0 46.0 L66.4 46.0 L67.8 46.0 L69.1 46.0 L70.5 46.0 L71.8 46.0 L73.2 46.0 L74.6 46.0 L75.9 46.0 L77.3 46.0 L78.6 46.0 L80.0 46.0 L81.4 46.0 L82.7 46.0 L84.1 46.0 L85.4 46.0 L86.8 46.0 L88.2 46.0 L89.5 46.0 L90.9 46.0 L92.2 46.0 L93.6 46.0 L95.0 46.0 L96.3 46.0 L97.7 46.0 L99.0 46.0 L100.4 46.0 L101.8 46.0 L103.1 46.0 L104.5 46.0 L105.8 46.0 L107.2 46.0 L108.6 46.0 L109.9 46.0 L111.3 46.0 L112.6 46.0 L114.0 46.0 L115.4 46.0 L116.7 46.0 L118.1 46.0 L119.4 46.0 L120.8 46.0 L122.2 46.0 L123.5 46.0 L124.9 46.0 L126.2 46.0 L127.6 46.0 L129.0 46.0 L130.3 46.0 L131.7 46.0 L133.0 46.0 L134.4 46.0 L135.8 46.0 L137.1 46.0 L138.5 46.0 L139.8 46.0 L141.2 46.0 L142.6 46.0 L143.9 46.0 L145.3 46.0 L146.6 46.0 L148.0 46.0"/><circle class="a-d" cx="12.0" cy="47.9" r="1.1"/><circle class="a-d" cx="21.4" cy="50.5" r="1.1"/><circle class="a-d" cx="30.8" cy="57.7" r="1.1"/><circle class="a-d" cx="40.2" cy="46.6" r="1.1"/><circle class="a-d" cx="49.6" cy="46.3" r="1.1"/><circle class="a-d" cx="59.0" cy="34.8" r="1.1"/><circle class="a-d" cx="68.4" cy="51.5" r="1.1"/><circle class="a-d" cx="77.8" cy="42.8" r="1.1"/><circle class="a-d" cx="87.2" cy="51.2" r="1.1"/><circle class="a-d" cx="96.6" cy="31.4" r="1.1"/><circle class="a-d" cx="106.0" cy="29.8" r="1.1"/><circle class="a-d" cx="115.4" cy="23.4" r="1.1"/><circle class="a-d" cx="124.8" cy="17.1" r="1.1"/><circle class="a-d" cx="134.2" cy="44.2" r="1.1"/><circle class="a-d" cx="143.6" cy="74.0" r="1.1"/></svg></span><span class="al-meta"><span class="al-tag">Large tabular foundation model</span><span>Advanced</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Probabilistic regression: the whole predictive distribution, scored properly</span><span class="al-d">See a tabular foundation model predict a full distribution: growing noise, two outcomes, calibrated intervals, NLL and CRPS against a Gaussian baseline.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/tfm-imputation" data-tag="large tabular foundation model" data-q="imputation: filling missing cells with a tabular foundation model, mice, knn and midas imputation"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-g" x="14.0" y="12.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="41.0" y="12.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="68.0" y="12.0" width="24.0" height="10.0" rx="1"/><rect class="a-f" x="95.0" y="12.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="122.0" y="12.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="14.0" y="25.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="41.0" y="25.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="68.0" y="25.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="95.0" y="25.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="122.0" y="25.0" width="24.0" height="10.0" rx="1"/><rect class="a-f" x="14.0" y="38.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="41.0" y="38.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="68.0" y="38.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="95.0" y="38.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="122.0" y="38.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="14.0" y="51.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="41.0" y="51.0" width="24.0" height="10.0" rx="1"/><rect class="a-f" x="68.0" y="51.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="95.0" y="51.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="122.0" y="51.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="14.0" y="64.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="41.0" y="64.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="68.0" y="64.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="95.0" y="64.0" width="24.0" height="10.0" rx="1"/><rect class="a-f" x="122.0" y="64.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="14.0" y="77.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="41.0" y="77.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="68.0" y="77.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="95.0" y="77.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="122.0" y="77.0" width="24.0" height="10.0" rx="1"/><text class="a-t" x="101.0" y="19.0">?</text><text class="a-t" x="20.0" y="45.0">?</text><text class="a-t" x="74.0" y="58.0">?</text><text class="a-t" x="128.0" y="71.0">?</text></svg></span><span class="al-meta"><span class="al-tag">Large tabular foundation model</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Imputation: filling missing cells with a tabular foundation model, MICE, kNN and MIDAS</span><span class="al-d">Hide cells under MCAR, MAR or MNAR and compare five imputers live: mean, kNN, MICE, a MIDAS autoencoder and a tabular foundation model.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/tfm-masked-value" data-tag="large tabular foundation model" data-q="masked-value prediction: the pre-training task of a tabular foundation model masked value"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-g" x="14.0" y="12.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="41.0" y="12.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="68.0" y="12.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="95.0" y="12.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="122.0" y="12.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="14.0" y="25.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="41.0" y="25.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="68.0" y="25.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="95.0" y="25.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="122.0" y="25.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="14.0" y="38.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="41.0" y="38.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="68.0" y="38.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="95.0" y="38.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="122.0" y="38.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="14.0" y="51.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="41.0" y="51.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="68.0" y="51.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="95.0" y="51.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="122.0" y="51.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="14.0" y="64.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="41.0" y="64.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="68.0" y="64.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="95.0" y="64.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="122.0" y="64.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="14.0" y="77.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="41.0" y="77.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="68.0" y="77.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="95.0" y="77.0" width="24.0" height="10.0" rx="1"/><rect class="a-g" x="122.0" y="77.0" width="24.0" height="10.0" rx="1"/><rect class="a-f" x="68.0" y="38.0" width="24.0" height="10.0" rx="1"/><path class="a-b" d="M80.0 43.0L26.0 17.0"/><path class="a-b" d="M80.0 43.0L134.0 17.0"/><path class="a-l2" d="M80.0 43.0L26.0 43.0"/><path class="a-l2" d="M80.0 43.0L134.0 43.0"/><path class="a-b" d="M80.0 43.0L26.0 69.0"/><path class="a-b" d="M80.0 43.0L134.0 69.0"/><path class="a-l2" d="M80.0 43.0L26.0 82.0"/><path class="a-l2" d="M80.0 43.0L134.0 82.0"/></svg></span><span class="al-meta"><span class="al-tag">Large tabular foundation model</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Masked-value prediction: the pre-training task of a tabular foundation model</span><span class="al-d">Mask any cell and see its predicted distribution, the rows the model read and whether the truth falls inside: the skill behind prediction and imputation.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/midas-imputation" data-tag="large tabular foundation model" data-q="midas: multiple imputation with a denoising autoencoder midas"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><path class="a-l2" d="M14.0 20.0 L60.0 30.0 L100.0 30.0 L146.0 20.0"/><path class="a-l2" d="M14.0 32.0 L60.0 36.0 L100.0 36.0 L146.0 32.0"/><path class="a-l2" d="M14.0 44.0 L60.0 42.0 L100.0 42.0 L146.0 44.0"/><path class="a-l2" d="M14.0 56.0 L60.0 48.0 L100.0 48.0 L146.0 56.0"/><path class="a-l2" d="M14.0 68.0 L60.0 54.0 L100.0 54.0 L146.0 68.0"/><circle class="a-d" cx="14.0" cy="22.0" r="1.6"/><circle class="a-d" cx="14.0" cy="40.0" r="1.6"/><circle class="a-d" cx="14.0" cy="58.0" r="1.6"/><circle class="a-d" cx="60.0" cy="22.0" r="1.6"/><circle class="a-d" cx="60.0" cy="40.0" r="1.6"/><circle class="a-d" cx="60.0" cy="58.0" r="1.6"/><circle class="a-db" cx="80.0" cy="22.0" r="1.6"/><circle class="a-db" cx="80.0" cy="40.0" r="1.6"/><circle class="a-db" cx="80.0" cy="58.0" r="1.6"/><circle class="a-d" cx="100.0" cy="22.0" r="1.6"/><circle class="a-d" cx="100.0" cy="40.0" r="1.6"/><circle class="a-d" cx="100.0" cy="58.0" r="1.6"/><circle class="a-d" cx="146.0" cy="22.0" r="1.6"/><circle class="a-d" cx="146.0" cy="40.0" r="1.6"/><circle class="a-d" cx="146.0" cy="58.0" r="1.6"/><circle class="a-db" cx="125.7" cy="78.0" r="1.2"/><circle class="a-db" cx="127.7" cy="78.0" r="1.2"/><circle class="a-db" cx="134.0" cy="78.0" r="1.2"/><circle class="a-db" cx="120.6" cy="78.0" r="1.2"/><circle class="a-db" cx="120.2" cy="78.0" r="1.2"/><circle class="a-db" cx="112.9" cy="78.0" r="1.2"/><circle class="a-db" cx="123.0" cy="78.0" r="1.2"/><circle class="a-db" cx="118.5" cy="78.0" r="1.2"/><circle class="a-db" cx="122.2" cy="78.0" r="1.2"/><circle class="a-db" cx="114.4" cy="78.0" r="1.2"/></svg></span><span class="al-meta"><span class="al-tag">Large tabular foundation model</span><span>Advanced</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">MIDAS: multiple imputation with a denoising autoencoder</span><span class="al-d">Train a denoising autoencoder on a table with gaps in your browser, draw m imputations with dropout, and see the uncertainty that a single imputation hides.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/missingness-shift-mirrams" data-tag="large tabular foundation model" data-q="missingness shift: robustness when test data go missing differently (mirrams) missingness shift"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><path class="a-l2" d="M10.0 20.0L150.0 20.0"/><path class="a-l2" d="M10.0 38.0L150.0 38.0"/><path class="a-l2" d="M10.0 56.0L150.0 56.0"/><path class="a-l2" d="M10.0 74.0L150.0 74.0"/><path class="a-b" d="M12.0 20.0 L25.6 20.6 L39.2 21.2 L52.8 21.8 L66.4 22.4 L80.0 23.0 L93.6 23.6 L107.2 24.2 L120.8 24.8 L134.4 25.4 L148.0 26.0"/><path class="a-l" d="M12.0 22.0 L25.6 24.2 L39.2 26.4 L52.8 28.6 L66.4 30.8 L80.0 33.0 L93.6 35.2 L107.2 37.4 L120.8 39.6 L134.4 41.8 L148.0 44.0"/><path class="a-l2" d="M12.0 24.0 L25.6 28.6 L39.2 33.2 L52.8 37.8 L66.4 42.4 L80.0 47.0 L93.6 51.6 L107.2 56.2 L120.8 60.8 L134.4 65.4 L148.0 70.0"/><path class="a-l2" d="M26.0 12.0L26.0 80.0"/></svg></span><span class="al-meta"><span class="al-tag">Large tabular foundation model</span><span>Advanced</span><span class="al-real">Real computation</span><span class="al-lab">Code lab</span></span><span class="al-t">Missingness shift: robustness when test data go missing differently (MIRRAMS)</span><span class="al-d">Train at one missing rate and test as missingness grows or changes mechanism, as in MIRRAMS: mean imputation, indicators, mask augmentation, in-context.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/tfm-large-tables" data-tag="large tabular foundation model" data-q="large tables: choosing the context when a table outgrows it large tables"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-g" x="12.0" y="12.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="19.0" y="12.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="26.0" y="12.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="33.0" y="12.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="40.0" y="12.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="47.0" y="12.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="54.0" y="12.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="61.0" y="12.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="68.0" y="12.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="75.0" y="12.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="12.0" y="19.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="19.0" y="19.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="26.0" y="19.0" width="5.5" height="5.5" rx="1"/><rect class="a-f" x="33.0" y="19.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="40.0" y="19.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="47.0" y="19.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="54.0" y="19.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="61.0" y="19.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="68.0" y="19.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="75.0" y="19.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="12.0" y="26.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="19.0" y="26.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="26.0" y="26.0" width="5.5" height="5.5" rx="1"/><rect class="a-f" x="33.0" y="26.0" width="5.5" height="5.5" rx="1"/><rect class="a-f" x="40.0" y="26.0" width="5.5" height="5.5" rx="1"/><rect class="a-f" x="47.0" y="26.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="54.0" y="26.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="61.0" y="26.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="68.0" y="26.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="75.0" y="26.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="12.0" y="33.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="19.0" y="33.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="26.0" y="33.0" width="5.5" height="5.5" rx="1"/><rect class="a-f" x="33.0" y="33.0" width="5.5" height="5.5" rx="1"/><rect class="a-f" x="40.0" y="33.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="47.0" y="33.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="54.0" y="33.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="61.0" y="33.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="68.0" y="33.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="75.0" y="33.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="12.0" y="40.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="19.0" y="40.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="26.0" y="40.0" width="5.5" height="5.5" rx="1"/><rect class="a-f" x="33.0" y="40.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="40.0" y="40.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="47.0" y="40.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="54.0" y="40.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="61.0" y="40.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="68.0" y="40.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="75.0" y="40.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="12.0" y="47.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="19.0" y="47.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="26.0" y="47.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="33.0" y="47.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="40.0" y="47.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="47.0" y="47.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="54.0" y="47.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="61.0" y="47.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="68.0" y="47.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="75.0" y="47.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="12.0" y="54.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="19.0" y="54.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="26.0" y="54.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="33.0" y="54.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="40.0" y="54.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="47.0" y="54.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="54.0" y="54.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="61.0" y="54.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="68.0" y="54.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="75.0" y="54.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="12.0" y="61.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="19.0" y="61.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="26.0" y="61.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="33.0" y="61.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="40.0" y="61.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="47.0" y="61.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="54.0" y="61.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="61.0" y="61.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="68.0" y="61.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="75.0" y="61.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="12.0" y="68.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="19.0" y="68.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="26.0" y="68.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="33.0" y="68.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="40.0" y="68.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="47.0" y="68.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="54.0" y="68.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="61.0" y="68.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="68.0" y="68.0" width="5.5" height="5.5" rx="1"/><rect class="a-g" x="75.0" y="68.0" width="5.5" height="5.5" rx="1"/><path class="a-b" d="M84.0 40.0L116.0 40.0"/><rect class="a-f" x="118.0" y="28.0" width="30.0" height="24.0" rx="3"/><circle class="a-db" cx="133.0" cy="40.0" r="2.6"/></svg></span><span class="al-meta"><span class="al-tag">Large tabular foundation model</span><span>Advanced</span><span class="al-real">Real computation</span><span class="al-lab">Code lab</span></span><span class="al-t">Large tables: choosing the context when a table outgrows it</span><span class="al-d">Give a tabular foundation model a 3,000-row table and a small context: compare random, class-balanced, ensembled and retrieved context rows at six budgets.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/tfm-any-table" data-tag="large tabular foundation model" data-q="any table, any schema: column order, names and units do not matter any table"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><rect class="a-g" x="12.0" y="18.0" width="40.0" height="46.0" rx="3"/><path class="a-l2" d="M16.0 26.0L48.0 26.0"/><path class="a-l2" d="M16.0 33.0L48.0 33.0"/><path class="a-l2" d="M16.0 40.0L48.0 40.0"/><path class="a-l2" d="M16.0 47.0L48.0 47.0"/><path class="a-l2" d="M16.0 54.0L48.0 54.0"/><rect class="a-g" x="60.0" y="24.0" width="40.0" height="46.0" rx="3"/><path class="a-l2" d="M64.0 32.0L96.0 32.0"/><path class="a-l2" d="M64.0 39.0L96.0 39.0"/><path class="a-l2" d="M64.0 46.0L96.0 46.0"/><path class="a-l2" d="M64.0 53.0L96.0 53.0"/><path class="a-l2" d="M64.0 60.0L96.0 60.0"/><rect class="a-g" x="108.0" y="30.0" width="40.0" height="46.0" rx="3"/><path class="a-l2" d="M112.0 38.0L144.0 38.0"/><path class="a-l2" d="M112.0 45.0L144.0 45.0"/><path class="a-l2" d="M112.0 52.0L144.0 52.0"/><path class="a-l2" d="M112.0 59.0L144.0 59.0"/><path class="a-l2" d="M112.0 66.0L144.0 66.0"/><path class="a-b" d="M32.0 72.0 L80.0 78.0 L128.0 84.0"/><circle class="a-db" cx="32.0" cy="72.0" r="2.0"/><circle class="a-db" cx="80.0" cy="78.0" r="2.0"/><circle class="a-db" cx="128.0" cy="84.0" r="2.0"/></svg></span><span class="al-meta"><span class="al-tag">Large tabular foundation model</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Any table, any schema: column order, names and units do not matter</span><span class="al-d">One predictor reads three unrelated tables with no retraining, and its predictions survive shuffled column order, renamed columns and changed units unchanged.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/tfm-calibration" data-tag="large tabular foundation model" data-q="calibration: do a tabular model&#x27;s probabilities mean what they say? calibration"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><path class="a-l2" d="M20.0 78.0L140.0 12.0"/><path class="a-b" d="M20.0 78.0 L32.0 74.8 L44.0 64.8 L56.0 58.8 L68.0 51.6 L80.0 41.2 L92.0 38.4 L104.0 34.4 L116.0 25.2 L128.0 20.2 L140.0 12.0"/><circle class="a-db" cx="20.0" cy="78.0" r="1.6"/><circle class="a-db" cx="32.0" cy="74.8" r="1.6"/><circle class="a-db" cx="44.0" cy="64.8" r="1.6"/><circle class="a-db" cx="56.0" cy="58.8" r="1.6"/><circle class="a-db" cx="68.0" cy="51.6" r="1.6"/><circle class="a-db" cx="80.0" cy="41.2" r="1.6"/><circle class="a-db" cx="92.0" cy="38.4" r="1.6"/><circle class="a-db" cx="104.0" cy="34.4" r="1.6"/><circle class="a-db" cx="116.0" cy="25.2" r="1.6"/><circle class="a-db" cx="128.0" cy="20.2" r="1.6"/><circle class="a-db" cx="140.0" cy="12.0" r="1.6"/><path class="a-l" d="M20.0 78.0 L32.0 57.1 L44.0 48.5 L56.0 41.9 L68.0 36.3 L80.0 31.3 L92.0 26.9 L104.0 22.8 L116.0 19.0 L128.0 15.4 L140.0 12.0"/></svg></span><span class="al-meta"><span class="al-tag">Large tabular foundation model</span><span>Intermediate</span><span class="al-real">Real computation</span><span class="al-lab">Code lab</span></span><span class="al-t">Calibration: do a tabular model&#x27;s probabilities mean what they say?</span><span class="al-d">Draw a reliability diagram for in-context and logistic predictions, measure ECE, Brier score and log loss, and fix overconfidence with temperature scaling.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a><a class="al-card" href="/ai-simulations/tfm-anomaly-detection" data-tag="large tabular foundation model" data-q="anomaly detection with masked-cell surprise anomaly detection"><span class="al-img"><svg viewBox="0 0 160 90" aria-hidden="true"><circle class="a-d" cx="84.9" cy="35.3" r="0.9"/><circle class="a-d" cx="82.0" cy="69.4" r="0.9"/><circle class="a-d" cx="95.4" cy="55.5" r="0.9"/><circle class="a-d" cx="56.9" cy="39.8" r="0.9"/><circle class="a-d" cx="95.5" cy="60.1" r="0.9"/><circle class="a-d" cx="44.9" cy="45.9" r="0.9"/><circle class="a-d" cx="124.5" cy="51.6" r="0.9"/><circle class="a-d" cx="95.4" cy="66.4" r="0.9"/><circle class="a-d" cx="45.8" cy="44.9" r="0.9"/><circle class="a-d" cx="86.5" cy="56.1" r="0.9"/><circle class="a-d" cx="106.7" cy="57.4" r="0.9"/><circle class="a-d" cx="68.1" cy="44.2" r="0.9"/><circle class="a-d" cx="82.6" cy="50.8" r="0.9"/><circle class="a-d" cx="87.1" cy="61.4" r="0.9"/><circle class="a-d" cx="54.0" cy="53.5" r="0.9"/><circle class="a-d" cx="110.5" cy="57.1" r="0.9"/><circle class="a-d" cx="109.3" cy="43.7" r="0.9"/><circle class="a-d" cx="78.5" cy="21.5" r="0.9"/><circle class="a-d" cx="96.6" cy="53.9" r="0.9"/><circle class="a-d" cx="107.6" cy="56.5" r="0.9"/><circle class="a-d" cx="71.7" cy="44.6" r="0.9"/><circle class="a-d" cx="50.7" cy="53.4" r="0.9"/><circle class="a-d" cx="71.3" cy="43.9" r="0.9"/><circle class="a-d" cx="85.7" cy="44.3" r="0.9"/><circle class="a-d" cx="106.9" cy="17.6" r="0.9"/><circle class="a-d" cx="80.3" cy="15.6" r="0.9"/><circle class="a-d" cx="67.6" cy="44.3" r="0.9"/><circle class="a-d" cx="51.0" cy="49.6" r="0.9"/><circle class="a-d" cx="82.1" cy="33.7" r="0.9"/><circle class="a-d" cx="84.5" cy="44.9" r="0.9"/><circle class="a-d" cx="74.9" cy="46.1" r="0.9"/><circle class="a-d" cx="45.9" cy="75.5" r="0.9"/><circle class="a-d" cx="118.7" cy="70.8" r="0.9"/><circle class="a-d" cx="102.7" cy="35.2" r="0.9"/><circle class="a-d" cx="99.1" cy="50.9" r="0.9"/><circle class="a-d" cx="77.1" cy="40.7" r="0.9"/><circle class="a-d" cx="77.6" cy="51.3" r="0.9"/><circle class="a-d" cx="102.2" cy="22.7" r="0.9"/><circle class="a-d" cx="106.6" cy="36.3" r="0.9"/><circle class="a-d" cx="74.3" cy="43.9" r="0.9"/><circle class="a-d" cx="49.8" cy="48.5" r="0.9"/><circle class="a-d" cx="80.2" cy="43.1" r="0.9"/><circle class="a-d" cx="87.7" cy="46.3" r="0.9"/><circle class="a-d" cx="87.9" cy="43.2" r="0.9"/><circle class="a-d" cx="64.2" cy="55.1" r="0.9"/><circle class="a-d" cx="87.2" cy="41.5" r="0.9"/><circle class="a-d" cx="77.8" cy="67.6" r="0.9"/><circle class="a-d" cx="89.2" cy="49.9" r="0.9"/><circle class="a-d" cx="124.8" cy="51.0" r="0.9"/><circle class="a-d" cx="82.2" cy="60.8" r="0.9"/><circle class="a-d" cx="43.1" cy="70.5" r="0.9"/><circle class="a-d" cx="74.2" cy="29.4" r="0.9"/><circle class="a-d" cx="65.8" cy="46.7" r="0.9"/><circle class="a-d" cx="84.5" cy="41.0" r="0.9"/><circle class="a-d" cx="69.1" cy="38.7" r="0.9"/><circle class="a-d" cx="76.7" cy="59.7" r="0.9"/><circle class="a-d" cx="89.0" cy="61.6" r="0.9"/><circle class="a-d" cx="83.2" cy="47.1" r="0.9"/><circle class="a-d" cx="84.9" cy="50.2" r="0.9"/><circle class="a-d" cx="95.5" cy="35.6" r="0.9"/><circle class="a-d" cx="66.5" cy="26.3" r="0.9"/><circle class="a-d" cx="102.4" cy="72.7" r="0.9"/><circle class="a-d" cx="65.1" cy="55.3" r="0.9"/><circle class="a-d" cx="52.3" cy="47.4" r="0.9"/><circle class="a-d" cx="81.4" cy="37.8" r="0.9"/><circle class="a-d" cx="89.0" cy="34.3" r="0.9"/><circle class="a-d" cx="79.8" cy="71.2" r="0.9"/><circle class="a-d" cx="104.7" cy="28.1" r="0.9"/><circle class="a-d" cx="62.2" cy="71.0" r="0.9"/><circle class="a-d" cx="87.7" cy="13.0" r="0.9"/><circle class="a-db" cx="24.0" cy="16.0" r="2.6"/><circle class="a-b" cx="24" cy="16" r="6"/><circle class="a-db" cx="140.0" cy="74.0" r="2.6"/><circle class="a-b" cx="140" cy="74" r="6"/><circle class="a-db" cx="130.0" cy="18.0" r="2.6"/><circle class="a-b" cx="130" cy="18" r="6"/></svg></span><span class="al-meta"><span class="al-tag">Large tabular foundation model</span><span>Advanced</span><span class="al-real">Real computation</span><span class="al-own">Your input</span><span class="al-lab">Code lab</span></span><span class="al-t">Anomaly detection with masked-cell surprise</span><span class="al-d">Plant anomalies in a table and find them without labels: every cell is predicted from the rest of its row, and improbable combinations rise to the top.</span><span class="al-go">Open simulation <i aria-hidden="true">→</i></span></a></div></section></div>

## About these simulations

Every simulation runs in your browser; nothing is sent to a server. Those marked *Real computation* calculate every number with the equations described on their page, on real or freshly generated data. Two agent simulations replay scripted runs with typical values; the agent and guardrail simulations marked *Real tools, rule-based planner* run real tools and decide with rules that stand in for a language model, so you can try your own questions without an API key. Every simulation has a code lab: the same idea in runnable Python, step by step. Open any card to read it in the simulation reader; use the arrows to move between simulations and the Code lab tab to switch to the code.
