A Closer Look at the “Societies of Thought” Paper
Published on February 12, 2026 9:38 PM GMTIntroductionToday I’m going to take a deep dive…
Category: Opinion
models have some pretty funny attractor states
Published on February 12, 2026 9:14 PM GMTThis work was conducted during the MATS 9.0…
Stay in your human loop
Published on February 12, 2026 9:05 PM GMTI drafted up a Slack message, copied the…
Multiverse sampling assumption
Published on February 12, 2026 7:59 PM GMTEpistemic: generally agree with S.Armstrong here, but adding…
What We Learned from Briefing 140+ Lawmakers on the Threat from AI
Published on February 12, 2026 7:53 PM GMTBack in May 2025, I published a post…
Paper: Prompt Optimization Makes Misalignment Legible
Published on February 12, 2026 7:45 PM GMT📄 Link to paper (preprint)We are sharing an…
Claude’s Constitution
Published on February 12, 2026 7:44 PM GMTTL;DR: Anthropic has made important progress at setting…
Human-like metacognitive skills will reduce LLM slop and aid alignment and capabilities
Published on February 12, 2026 7:38 PM GMT1. Summary and overviewLLMs seem to lack metacognitive…
Research note: A simpler AI timelines model predicts 99% AI R&D automation in ~2032
Published on February 12, 2026 12:13 AM GMTIn this post, I describe a simple model…
[Paper] How does information access affect LLM monitors’ ability to detect sabotage?
Published on February 11, 2026 9:25 PM GMTTL;DRWe evaluate LLM monitors in three AI control…