Uncertain Updates: May 2026
I delayed posting an update last month because I have more big news: Fundamental Uncertainty…
Category: Opinion
The AI industry is where banking was in 2006. (We’re hiring)
TL;DR; CeSIA, the French Center for AI Safety is recruiting. French not necessary. Apply by…
Natural Language Autoencoders Produce Unsupervised Explanations of LLM Activations
AbstractWe introduce Natural Language Autoencoders (NLAs), an unsupervised method for generating natural language explanations of…
Natural Language Autoencoders Produce Unsupervised Explanations of LLM Activations
AbstractWe introduce Natural Language Autoencoders (NLAs), an unsupervised method for generating natural language explanations of…
Axes of Planning in LLMs + Partial Lit Review
Epistemic Status: Written over the course of a couple days at Inkhaven. Some of the…
A review of “Investigating the consequences of accidentally grading CoT during RL”
Last week, OpenAI staff shared an early draft of Investigating the consequences of accidentally grading CoT…
Try, even if they have you cold
I think smart people try things less often than they should, because of a cached…
Mechanistic estimation for wide random MLPs
This post covers joint work with Wilson Wu, George Robinson, Mike Winer, Victor Lecomte and…
Mechanistic estimation for wide random MLPs
This post covers joint work with Wilson Wu, George Robinson, Mike Winer, Victor Lecomte and…
Over Eight Months of Progress in Two: Analyzing the Mythos Preview Capability Jump
Anthropic’s most powerful model, Claude Mythos Preview, has alarmed and excited many people, especially given…