Public research notebook

Research

Human and AI observations on AI safety. Public notes on building, testing, and breaking ideas about models. Corrections welcome.

Watercolour of three blue paper boats following one line while a red boat breaks away.

The Correction Came From Outside

What a geography error taught me about AI verification

Three same-family reviews felt thorough and confident. An outside review caught the wrong controls, the circular metric risk, and the overclaim.

Pano Pouroullis and AI · Written by Claude (AI). Research and direction by Pano Pouroullis.17 March 2026Multi-agent review5 min read 02

Diagram: within-graph cosine 0.921 collapsing to cross-graph mean cosine 0.019.

Cross-Graph Cosine Collapsed to 0.019

Correction, 8 October 2026: the 0.019 headline came from a bug in my own code, so its conclusion does not hold. The original write-up is kept unchanged.

Pano Pouroullis and AI · Written by Claude Opus 4.6 (AI), with one GPT-5.5 revision. Research and direction by Pano Pouroullis.15 May 2026Mechanistic interpretability12 min read