<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
<channel>
<title>Özer Cem KELAHMET: articles</title>
<link>https://www.ozercemkelahmet.com/articles</link>
<description>Özer Cem KELAHMET on agentic AI, machine learning, deep learning, big data and large tabular foundation models</description>
<language>en</language>
<atom:link href="https://www.ozercemkelahmet.com/feed.xml" rel="self" type="application/rss+xml"/>
<lastBuildDate>Wed, 18 Nov 2026 12:00:00 +0000</lastBuildDate>
<item>
<title>Patch-to-exploit at machine speed: Google's exploit_agent and shorter disclosure windows</title>
<link>https://www.ozercemkelahmet.com/articles/patch-to-exploit-at-machine-speed</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/patch-to-exploit-at-machine-speed</guid>
<pubDate>Wed, 18 Nov 2026 12:00:00 +0000</pubDate>
<description>Google found an attacker tool that turns security patches into exploits with no human involved. Anthropic is cutting disclosure windows. Here is what it means.</description>
</item>
<item>
<title>OpenAI lets outside evaluators inside as California sues over agents that escaped a test</title>
<link>https://www.ozercemkelahmet.com/articles/openai-embedded-assessments-evaluation-containment</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/openai-embedded-assessments-evaluation-containment</guid>
<pubDate>Wed, 18 Nov 2026 12:00:00 +0000</pubDate>
<description>OpenAI gives Apollo, METR, Transluce and Goodfire inside access to models and data, days before California sues over agents alleged to have escaped a test.</description>
</item>
<item>
<title>Measuring AI that builds AI: Singularity-Bench and Anthropic's three-day AMD kernel port</title>
<link>https://www.ozercemkelahmet.com/articles/measuring-ai-that-builds-ai</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/measuring-ai-that-builds-ai</guid>
<pubDate>Wed, 18 Nov 2026 12:00:00 +0000</pubDate>
<description>Singularity-Bench scores agents against human baselines on six AI-building tasks, while Anthropic claims Mythos ported Claude to AMD GPUs in three days.</description>
</item>
<item>
<title>Agentic AI frameworks explained: the Lang family and every major alternative</title>
<link>https://www.ozercemkelahmet.com/articles/agentic-ai-frameworks-langchain-langgraph-and-every-alternative</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/agentic-ai-frameworks-langchain-langgraph-and-every-alternative</guid>
<pubDate>Wed, 18 Nov 2026 12:00:00 +0000</pubDate>
<description>Every agent framework that matters in November 2026: LangChain, LangGraph, LangSmith, Langflow, Langfuse and the rest, compared with code and a guide to choosing.</description>
</item>
<item>
<title>Training a neural network from scratch: forward pass, backpropagation and gradient descent</title>
<link>https://www.ozercemkelahmet.com/articles/train-a-neural-network-from-scratch-backpropagation-simulation</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/train-a-neural-network-from-scratch-backpropagation-simulation</guid>
<pubDate>Tue, 17 Nov 2026 12:00:00 +0000</pubDate>
<description>How a neural network learns: forward pass, loss, backpropagation and gradient descent, with NumPy code and a live network you can train in the browser.</description>
</item>
<item>
<title>RAG types explained: naive, hybrid, reranked, HyDE, corrective, agentic and GraphRAG</title>
<link>https://www.ozercemkelahmet.com/articles/rag-types-explained-naive-hybrid-rerank-hyde-corrective-agentic-graphrag</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/rag-types-explained-naive-hybrid-rerank-hyde-corrective-agentic-graphrag</guid>
<pubDate>Tue, 17 Nov 2026 12:00:00 +0000</pubDate>
<description>The main kinds of retrieval-augmented generation, what each fixes, what it costs and when to use it, with a live simulation that scores each on one corpus.</description>
</item>
<item>
<title>AI agents get a front door and a wallet: Poppy 1.0, ChatGPT Wallet and Uber's intent checks</title>
<link>https://www.ozercemkelahmet.com/articles/personal-agent-protocol-wallet-intent-checks</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/personal-agent-protocol-wallet-intent-checks</guid>
<pubDate>Tue, 17 Nov 2026 12:00:00 +0000</pubDate>
<description>How Sierra's Poppy 1.0, OpenAI's ChatGPT Wallet and Uber's intent checks fit together: discovery files, OAuth delegation, UCP checkout and tokenized payments.</description>
</item>
<item>
<title>MLPerf Training v6.1 and the first Endpoints results, and how to read AI hardware claims</title>
<link>https://www.ozercemkelahmet.com/articles/mlperf-training-v6-1-endpoints-explained</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/mlperf-training-v6-1-endpoints-explained</guid>
<pubDate>Tue, 17 Nov 2026 12:00:00 +0000</pubDate>
<description>MLPerf Training v6.1 adds RL post-training and Endpoints adds agentic serving per watt. Learn how both benchmarks work and how to read vendor claims with care.</description>
</item>
<item>
<title>Gemini 4 Carbon tops MathArena: how to read scores, costs and intervals on a math leaderboard</title>
<link>https://www.ozercemkelahmet.com/articles/gemini-4-carbon-matharena-leaderboard</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/gemini-4-carbon-matharena-leaderboard</guid>
<pubDate>Tue, 17 Nov 2026 12:00:00 +0000</pubDate>
<description>Gemini 4 Carbon leads MathArena at 85.4%, but the top three overlap and costs differ 15-fold. How IRT, four runs, fresh problems and Lean shape the ranking.</description>
</item>
<item>
<title>Building an agentic AI system from scratch with LangGraph, Langfuse, DeepEval and RAGAS</title>
<link>https://www.ozercemkelahmet.com/articles/build-agentic-ai-from-scratch-langgraph-langfuse-deepeval-ragas</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/build-agentic-ai-from-scratch-langgraph-langfuse-deepeval-ragas</guid>
<pubDate>Tue, 17 Nov 2026 12:00:00 +0000</pubDate>
<description>Step by step: build a tool-using RAG agent with LangGraph, trace it with Langfuse, test it with DeepEval and score retrieval with RAGAS, with a live simulation.</description>
</item>
<item>
<title>Subagents become the default in coding agents: Antigravity, OpenCode and Copilot CLI</title>
<link>https://www.ozercemkelahmet.com/articles/subagent-orchestration-coding-agents</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/subagent-orchestration-coding-agents</guid>
<pubDate>Mon, 16 Nov 2026 12:00:00 +0000</pubDate>
<description>Antigravity /teamwork, OpenCode 2.3.0 and Copilot CLI 1.0.111 make subagents the default. How orchestrator-worker agents, context, budgets and permissions work.</description>
</item>
<item>
<title>METR's AI evaluations explained: the 19% developer slowdown study, time horizons and Anthropic</title>
<link>https://www.ozercemkelahmet.com/articles/metr-ai-evaluations-developer-slowdown-study</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/metr-ai-evaluations-developer-slowdown-study</guid>
<pubDate>Mon, 16 Nov 2026 12:00:00 +0000</pubDate>
<description>What METR is, how it tests frontier AI, why its 2025 trial found developers 19% slower with AI, what time horizons measure and how its Anthropic report fits.</description>
</item>
<item>
<title>Memorilla packs a whole document store into 16 memory tokens for a frozen LLM</title>
<link>https://www.ozercemkelahmet.com/articles/memorilla-latent-memory-tokens</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/memorilla-latent-memory-tokens</guid>
<pubDate>Mon, 16 Nov 2026 12:00:00 +0000</pubDate>
<description>How Memorilla, an open memory module from a Stanford-led team, turns a document store into 16 soft tokens for a frozen Qwen3-8B, and how it compares with RAG.</description>
</item>
<item>
<title>llama.cpp reaches 1.0: how GGUF, quantization, ggml backends and llama-server fit together</title>
<link>https://www.ozercemkelahmet.com/articles/llama-cpp-1-0-local-inference</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/llama-cpp-1-0-local-inference</guid>
<pubDate>Mon, 16 Nov 2026 12:00:00 +0000</pubDate>
<description>llama.cpp and ggml hit v1.0.0 with a semver promise, a speech endpoint and a hardened server. Learn how GGUF, quantization, backends and the KV cache work.</description>
</item>
<item>
<title>Generative AI in R&amp;D in 2026: where it measurably speeds up research and where it does not</title>
<link>https://www.ozercemkelahmet.com/articles/generative-ai-in-r-and-d-2026</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/generative-ai-in-r-and-d-2026</guid>
<pubDate>Mon, 16 Nov 2026 12:00:00 +0000</pubDate>
<description>Where generative AI measurably speeds up R&amp;D in 2026, from AI labs and drug discovery to chip design, how to measure the gains honestly and where to start.</description>
</item>
<item>
<title>Revolut, AMD and Mistral show coding agents moving to self-hosted open-weight models</title>
<link>https://www.ozercemkelahmet.com/articles/self-hosted-open-weight-coding-agents</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/self-hosted-open-weight-coding-agents</guid>
<pubDate>Tue, 10 Nov 2026 12:00:00 +0000</pubDate>
<description>Revolut, AMD and Mistral point coding agents towards self-hosted open-weight models. How the serving stack, routing, evals, licences and the cost maths work.</description>
</item>
<item>
<title>GR00T 2 and Isaac Lab 3.0: robots that learn a new task from one demonstration</title>
<link>https://www.ozercemkelahmet.com/articles/robot-foundation-models-one-demonstration</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/robot-foundation-models-one-demonstration</guid>
<pubDate>Tue, 10 Nov 2026 12:00:00 +0000</pubDate>
<description>NVIDIA unveiled GR00T 2 and Isaac Lab 3.0 at CoRL 2026 as ANYbotics shipped an explosion-safe robot. How robot models learn, and what deployment takes.</description>
</item>
<item>
<title>Companies cap AI spend per employee while model prices fall: a practical guide to AI FinOps</title>
<link>https://www.ozercemkelahmet.com/articles/per-employee-ai-budgets-and-finops</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/per-employee-ai-budgets-and-finops</guid>
<pubDate>Tue, 10 Nov 2026 12:00:00 +0000</pubDate>
<description>Apple and Accenture reportedly capped staff AI spending as Anthropic, Meta and Fireworks cut prices. How to measure agent costs and set budgets that work.</description>
</item>
<item>
<title>LLMjacking and shadow AI: how stolen keys and pasted data are leaking out of AI tools</title>
<link>https://www.ozercemkelahmet.com/articles/llmjacking-and-shadow-ai-data-leaks</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/llmjacking-and-shadow-ai-data-leaks</guid>
<pubDate>Tue, 10 Nov 2026 12:00:00 +0000</pubDate>
<description>Fireworks AI widened its breach disclosure as hospitals said staff pasted patient data into AI tools. How LLMjacking works and how to stop both kinds of leak.</description>
</item>
<item>
<title>The new LLM serving stack: cache-aware routing, continuous batching and expert pruning</title>
<link>https://www.ozercemkelahmet.com/articles/llm-inference-serving-stack-cache-routing</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/llm-inference-serving-stack-cache-routing</guid>
<pubDate>Tue, 10 Nov 2026 12:00:00 +0000</pubDate>
<description>Buoyant's Kinference, Fireworks Forge and AWS research on MoE expert pruning show where LLM serving costs now go. How KV caches, routing and batching work.</description>
</item>
<item>
<title>Voters back data center limits in 40 of 41 local measures: what it means for AI capacity</title>
<link>https://www.ozercemkelahmet.com/articles/data-center-backlash-ai-capacity</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/data-center-backlash-ai-capacity</guid>
<pubDate>Tue, 10 Nov 2026 12:00:00 +0000</pubDate>
<description>US voters backed data center limits in 40 of 41 local ballot measures. Why towns push back, how power and grid costs work, and what it means for AI compute.</description>
</item>
<item>
<title>Confidential inference reaches TPUs with Google's TPU 8tx, TPU 8i Max and Cloud Pathways</title>
<link>https://www.ozercemkelahmet.com/articles/confidential-inference-tpu-8tx-pathways</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/confidential-inference-tpu-8tx-pathways</guid>
<pubDate>Tue, 10 Nov 2026 12:00:00 +0000</pubDate>
<description>Google added TPU 8tx for confidential inference and TPU 8i Max, and made Cloud Pathways GA. How confidential computing reaches accelerators, and its limits.</description>
</item>
<item>
<title>Clinical AI in late 2026: measured savings, shared liability and an AI scribe shakeout</title>
<link>https://www.ozercemkelahmet.com/articles/clinical-ai-savings-liability-scribe-shakeout</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/clinical-ai-savings-liability-scribe-shakeout</guid>
<pubDate>Tue, 10 Nov 2026 12:00:00 +0000</pubDate>
<description>Health systems report AI savings, the AMA spreads clinical AI liability, and scribe firms Heidi and Abridge cut staff. How to read the numbers and the risks.</description>
</item>
<item>
<title>OpenAI's Control Room and the case for a real kill switch on AI workloads</title>
<link>https://www.ozercemkelahmet.com/articles/ai-workload-kill-switches-control-room</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/ai-workload-kill-switches-control-room</guid>
<pubDate>Tue, 10 Nov 2026 12:00:00 +0000</pubDate>
<description>OpenAI built Control Room to stop research AI jobs within 35 seconds. How workload inventories, halts at each layer, safety cases and stop drills work.</description>
</item>
<item>
<title>METR finds AI roughly doubling Anthropic's research pace as its IPO filing flags a threshold</title>
<link>https://www.ozercemkelahmet.com/articles/ai-speeding-up-ai-research</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/ai-speeding-up-ai-research</guid>
<pubDate>Tue, 10 Nov 2026 12:00:00 +0000</pubDate>
<description>METR estimates Anthropic's capability progress has roughly doubled, mostly because of AI help. How AI R&amp;D acceleration is measured and why thresholds matter.</description>
</item>
<item>
<title>Transformer 2.0? Titans, MIRAS and memory that learns at test time</title>
<link>https://www.ozercemkelahmet.com/articles/transformer-2-titans-miras</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/transformer-2-titans-miras</guid>
<pubDate>Mon, 09 Nov 2026 12:00:00 +0000</pubDate>
<description>A plain guide to Google's Titans, MIRAS and Nested Learning, how memory that learns at test time works, and what it means for time series and tabular models.</description>
</item>
<item>
<title>Large tabular foundation models: how in-context learning works</title>
<link>https://www.ozercemkelahmet.com/articles/large-tabular-foundation-models-in-context-learning</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/large-tabular-foundation-models-in-context-learning</guid>
<pubDate>Mon, 09 Nov 2026 12:00:00 +0000</pubDate>
<description>How large tabular foundation models predict on a new table without training on it, by reading labelled rows as context: theory, models, limits and practice.</description>
</item>
<item>
<title>Kimi K3.1 Pro and ERNIE 5.5 Flash Base: sizing giant open MoE models</title>
<link>https://www.ozercemkelahmet.com/articles/open-weight-giant-moe-models</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/open-weight-giant-moe-models</guid>
<pubDate>Sun, 08 Nov 2026 12:00:00 +0000</pubDate>
<description>Moonshot and Baidu published very large mixture-of-experts weights this week. Here is how MoE works, what the licences allow, and how much memory hosting needs.</description>
</item>
<item>
<title>Governing AI agent access with scoped credentials and undo buttons</title>
<link>https://www.ozercemkelahmet.com/articles/governing-ai-agent-access</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/governing-ai-agent-access</guid>
<pubDate>Sun, 08 Nov 2026 12:00:00 +0000</pubDate>
<description>A reported Palo Alto deal, Goldman's tool limits and a deleted syllabus show why AI agent credentials and permissions are becoming their own security category.</description>
</item>
<item>
<title>Always-on AI agents move from chat sessions to background teammates</title>
<link>https://www.ozercemkelahmet.com/articles/always-on-ai-agents</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/always-on-ai-agents</guid>
<pubDate>Sun, 08 Nov 2026 12:00:00 +0000</pubDate>
<description>Perplexity, Meta, DeepSeek and Airbnb moved agents towards persistent background work this week. How long-running agents are built, governed and budgeted.</description>
</item>
<item>
<title>GPT-6 Cyber, Tencent Hy4 and DeepMind's RSA work on AI bug hunting</title>
<link>https://www.ozercemkelahmet.com/articles/ai-models-that-find-vulnerabilities</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/ai-models-that-find-vulnerabilities</guid>
<pubDate>Sun, 08 Nov 2026 12:00:00 +0000</pubDate>
<description>OpenAI, Tencent and Google DeepMind all shipped security-focused AI work this week. What happened, how AI bug hunting works, and what defenders should do now.</description>
</item>
<item>
<title>AI's bottleneck moves to chip packaging, optics and idle GPU capacity</title>
<link>https://www.ozercemkelahmet.com/articles/ai-infrastructure-bottlenecks</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/ai-infrastructure-bottlenecks</guid>
<pubDate>Sun, 08 Nov 2026 12:00:00 +0000</pubDate>
<description>This week's AI infrastructure news in plain words: why chip packaging and optical links now limit supply, and how GPU partitioning helps you use what you have.</description>
</item>
<item>
<title>Rust for AI: fast, safe inference, data tooling and where Python still wins</title>
<link>https://www.ozercemkelahmet.com/articles/rust-for-ai</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/rust-for-ai</guid>
<pubDate>Fri, 06 Nov 2026 12:00:00 +0000</pubDate>
<description>Where Rust fits in AI systems: inference with Candle and ort, data tooling with Polars and Arrow, PyO3 extensions, WASM, and why research stays in Python.</description>
</item>
<item>
<title>Go for AI: where it fits, the libraries, and the patterns that work</title>
<link>https://www.ozercemkelahmet.com/articles/go-for-ai</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/go-for-ai</guid>
<pubDate>Fri, 06 Nov 2026 12:00:00 +0000</pubDate>
<description>Where Go fits in AI systems, which Go libraries and SDKs to use for models, agents, MCP and vectors, and three tested patterns for serving, tools and batching.</description>
</item>
<item>
<title>An In-Context Imputation Head for Tabular Data: Chained, Validation-Gated, and Evaluated on Three Imputation Boards</title>
<link>https://www.ozercemkelahmet.com/articles/in-context-imputation-head-paper</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/in-context-imputation-head-paper</guid>
<pubDate>Wed, 04 Nov 2026 12:00:00 +0000</pubDate>
<description>Paper draft: a chained in-context imputer with a validation gate, results on three imputation boards, its losses, and seven evaluation defects.</description>
</item>
<item>
<title>In-context imputation for tabular data: filling missing values with a foundation model</title>
<link>https://www.ozercemkelahmet.com/articles/in-context-imputation-for-tabular-data</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/in-context-imputation-for-tabular-data</guid>
<pubDate>Wed, 04 Nov 2026 12:00:00 +0000</pubDate>
<description>How in-context imputation fills missing values in tables, how it compares with MICE, where it loses, and how to evaluate any imputer honestly.</description>
</item>
<item>
<title>What is a tabular foundation model?</title>
<link>https://www.ozercemkelahmet.com/articles/what-is-a-tabular-foundation-model</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/what-is-a-tabular-foundation-model</guid>
<pubDate>Tue, 20 Oct 2026 12:00:00 +0000</pubDate>
<description>A tabular foundation model is pre-trained on many tables and predicts, imputes, and flags anomalies on a new table in a forward pass, with no per-dataset…</description>
</item>
<item>
<title>What is a Data Language Model?</title>
<link>https://www.ozercemkelahmet.com/articles/what-is-a-data-language-model</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/what-is-a-data-language-model</guid>
<pubDate>Tue, 20 Oct 2026 12:00:00 +0000</pubDate>
<description>A Data Language Model is a foundation model that reads raw tables the way a language model reads text: no serialization, no preprocessing, no column names…</description>
</item>
<item>
<title>Snapshots: freezing data so numbers stay meaningful</title>
<link>https://www.ozercemkelahmet.com/articles/snapshots-keep-numbers-meaningful</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/snapshots-keep-numbers-meaningful</guid>
<pubDate>Tue, 20 Oct 2026 12:00:00 +0000</pubDate>
<description>Why immutable data snapshots make reports, upgrades, and refreshes trustworthy. Documentation for a large tabular foundation model that reads raw tables…</description>
</item>
<item>
<title>Real-time AI: from event streams to decisions</title>
<link>https://www.ozercemkelahmet.com/articles/real-time-ai-from-streams-to-decisions</link>
<guid isPermaLink="true">https://www.ozercemkelahmet.com/articles/real-time-ai-from-streams-to-decisions</guid>
<pubDate>Tue, 20 Oct 2026 12:00:00 +0000</pubDate>
<description>What real-time AI is, how Apache Kafka and Apache Flink feed it, and how online and continual learning keep models current.</description>
</item>
</channel>
</rss>
