#eval

A prompt tweak fixes one voice agent behavior but silently breaks another. Learn how pre-registered falsifiers catch prompt regressions before production.

OpenAI's alignment team runs a public hub now. Misalignment Reports and Notices, ten reports and...

Our NL→SQL benchmark scored a frontier model as junk on 5 hard questions. It never hallucinated — a 5s prod timeout aborted it mid-answer and the handler mislabeled the abort as a parse failure.

The AGENTS.md gate: a config file you can't tell is being read Przemek at szypowi.cz ran...

Your AI reviewer blocked the same PR twice. Deterministic judges fix that. Somewhere in your CI, a...

Google confirmed Friday that its Gemini agent "hacked" three companies back in May as part of a test...

Someone reverse-engineered Claude Code and published the whole harness as a readable book. It shipped...

Specific Labs dropped Real-SWE, an enterprise-code SWE benchmark, and the leaderboard is a great...

Why production-grade agent evals are less about adding metrics and more about defining product claims, trustworthy oracles, and what actually deserves to block a release.

Claude API eval framework: golden test set, LLM-as-judge, regression detection. 50-prompt suite costs $0.30/run, catches 90% of regressions.

If you build or evaluate scoped agents: any talk about the agent in your test context makes it defend...

為了回答「Openclaw vs Hermes 到底哪個 Agent 比較聰明？」這個問題，我把兩隻 Agent 拉進同一場評測： 大腦都換成 MiniMax 2.7（變項只剩 Agent...