# AI weekly: 19–25 October 2026

Every article published from 19–25 October 2026, newest first, with a short summary of each.

## This week's articles

### [Why build a large tabular foundation model?](/articles/large-tabular-foundation-model)

20 October · Concepts

Why build a large tabular foundation model? Scaling parameters, pre-training data and tasks into one checkpoint that reads raw tables across every domain.

### [What is a tabular foundation model?](/articles/what-is-a-tabular-foundation-model)

20 October · Concepts

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…

### [What is a Data Language Model?](/articles/what-is-a-data-language-model)

20 October · Concepts

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…

### [Snapshots: freezing data so numbers stay meaningful](/articles/snapshots-keep-numbers-meaningful)

20 October · Architecture

Why immutable data snapshots make reports, upgrades, and refreshes trustworthy. Documentation for a large tabular foundation model that reads raw tables…

### [Retrieval-augmented generation (RAG): from naive to agentic](/articles/rag-from-naive-to-agentic)

20 October · Concepts

How retrieval-augmented generation works, why naive RAG fails, and how advanced, graph, and agentic RAG fix it, with evaluation.

### [Real-time AI: from event streams to decisions](/articles/real-time-ai-from-streams-to-decisions)

20 October · Systems

What real-time AI is, how Apache Kafka and Apache Flink feed it, and how online and continual learning keep models current.

### [Quick runs and endpoints: stateless and stateful](/articles/quick-runs-and-endpoints)

20 October · Architecture

Two ways to use a table model: a one-shot call that stores nothing, and a persistent endpoint that holds state.

### [One pass, one bundle: what a table model returns](/articles/one-pass-six-outputs)

20 October · Concepts

One pass, one bundle: what a table model returns in a single call: sector, column profile, cross-table map, imputation, predictions and a held-out block.

### [Joining tables that share no key](/articles/joining-tables-without-a-key)

20 October · Concepts

Joining tables that share no key: how a model aligns columns and matches rows across sources, and why the result should be a reviewable proposal.

### [How a tabular foundation model took first place in probabilistic regression](/articles/probabilistic-regression-first-place)

20 October · Benchmarks

The levers that mattered, the mistakes that fooled us, and the price: a first on ScoringBench in CRPS and CRLS, 102 datasets and 50 rival models.

### [AI and machine learning for tabular data: a practical guide](/articles/ai-and-machine-learning-for-tables)

20 October · Concepts

How AI, machine learning, and deep learning relate, which learning paradigms apply to tables, and when to choose trees or a foundation model.

### [A score is only as good as its report](/articles/a-score-is-only-as-good-as-its-report)

20 October · Evaluation

A score is only as good as its report: what a held-out evaluation report should contain, and how to read one without being fooled by a headline number.

## More weeks

[Week 45, 2026 →](/weekly/2026-w45) · [All weeks](/weekly)
