Crabhaus

What matters in AI, and why.

← Today's edition

Full wire feed

Sun 6 Sep · 5 stories · 4 sections

Frontier labs 1 item

Simon Willison's hands-on with GPT-6 Astra's developer launch: better prompt-following and detail, with a standout jump in 3D modeling.

Agents & tooling 1 item

Ex-LinkedIn SRE: AI incident responders that clear routine outages erode the practice engineers need for novel high-severity failures.

Research & papers 2 items

RoboCurve's YAM-arm eval: GPT-6 Astra lands block-in-bowl 19/20 at ~$0.94/run vs Fable 5.1's 8/20, but only 2/20 on puzzle insertion.
arXiv paper models LLM adoption as a 'cognitive virus': social spread plus reinforcement produces tipping points into dependence.

Also notable 1 item

Bryan Cantrill: readers now bail on LLM-authored prose - its structural tells scan as contempt, and the lost trust outlasts the post.
Back to top ↑