Skip to content
The Wire

What happened, and whether it matters.

AI moves faster than it changes. Short notes on what landed this week, what it actually means, and a link to the source so you can disagree with me.

AllResearchProductPolicyFundingInfrastructureOpinion

July 2026

Jul 23, 2026

ResearchPublication

A frontier lab published its eval methodology instead of just the scores

The write-up covers the harness, the sampling temperature and the seeds — the three things that make a benchmark reproducible and the three things nobody publishes. Worth reading as a template for your own internal evals, not for the leaderboard position.

Abstract grayscale cover for: A frontier lab published its eval methodology instead of just the scores

Jul 13, 2026

ResearchResearch

The agent benchmark everyone quoted last quarter had a contamination problem

The task set overlapped with a public repository that appeared in pretraining data. Scores dropped several points on the cleaned set. A useful reminder that an agent benchmark is a software supply chain, not a measurement.

Every item links to its original source, which remains the work of its publisher. Headlines and summaries here are my own paraphrase and commentary, not reproductions.