Prompted CoT Early Exit Undermines the Monitoring Benefits of CoT Uncontrollability
Code: github.com/ElleNajt/controllability tldr: Yueh-Han et al. (2026) showed that models have a harder time making…
Category: Opinion
Arguments Should Be Decisive Criticisms
Critical Fallibilism (CF) is a philosophy I developed which deals with rationality, knowledge, and critical…
Humane Pesticides Are Massively Morally Urgent
Crosspost. I have lots of radical views about insects. I think probably most expected happiness…
“Best humans still outperform”: One turning point in the history of cope around artificial intelligence
A few years ago I was tickled by an article headline in a serious academic…
Against Doom & Pause AI
It is sometimes claimed that sufficiently advanced AI will almost certainly ("inevitably") kill everyone ("doom"),…
Come to Manifest 2026! (June 12-14)
Manifest is back! We're returning to Lighthaven on June 12-14! You can buy your tickets…
How Big Tech Becomes Ungovernable
Abstract: This post is an introduction to a concept I call "tech extensity"—when a company,…
Attempting to Quantify Chinese Bias in Open-Source LLMs
Over the past few years, Chinese LLMs have come to dominate in the sea of…
A Research Bet on SAE-like Expert Architectures
Interpretable by Construction: A Research Bet on SAE-like Expert Architectures The Bet You can build…
Church Planting: Lessons from the Comments
Last summer I got nerdsniped by evangelical christianity, and in particular church planting, the domestic…