Best of the last 7 days.
- When Your Prompt Gives Two Instructions That Collide: Write the Precedence Rule
- Slice Your Eval Results: An Average Score Hides the Users Who Got Worse
- Evaluating Summaries by Claims, Not Vibes: Separating Faithfulness From Coverage
- Golden-File Tests for Rendered Prompts: Catching Template Changes Before the Model Does
- Versioning Stored LLM Output: Schema Stamps, Upcasters, and Deliberate Backfills
- Pairwise Judging Needs Order Swaps: Controlling Position Bias in Prompt Comparisons
- 'Not Stated' Is an Answer: Designing Extraction Prompts That Don't Invent Missing Fields
- Trajectory Tests for Agents: Assert Invariants, Not Exact Steps
- Introducing Mistral Large 4 | Mistral — AI Hub + 282 more
- EmbeddingGemma 2: an open, lightweight multimodal embedding model — AI Hub + 190 more
- 2026 Climate Tech Companies to Watch — MIT Technology Review + 138 more
- ttok 0.4 — Simon Willison + 115 more
- Introducing Mistral Large 4: Le chonk — Simon Willison + 87 more
- Ex-Anthropic Researcher Coxon to Testify at NYC Hearing on AI — AI Hub + 50 more
- An Anthropic AI model sent a false homicide tip to Philadelphia police — TechCrunch — AI + 46 more