Anthropic: “Statement from Dario Amodei on our discussions with the Department of War”
I believe deeply in the existential importance of using AI to defend the United States…
Category: Opinion
Asymmetric Risks of Unfaithful Reasoning: Omission as the Critical Failure Mode for AI Monitoring
TLDR:Faithful reasoning is a representation in a comprehensible lower dimension that is understandable and only…
Getting Back To It
Artist: Lily TaylorIt’s been a while since I’ve written anything lately, and that doesn’t feel…
Inference-time Generative Debates on Coding and Reasoning Tasks for Scalable Oversight
By Ethan Elasky and Frank Nakasako (equal contribution)We tested generative debate (where participants freely make…
A minor point about instrumental convergence that I would like feedback on
PreambleMy current understanding: the EY/MIRI perspective is that superintelligent AI will invariably instrumentally converge on…
AI welfare as a demotivator for takeover.
TLDR: Superhuman AI may consider takeover the risky option, and we can influence its choices…
Frontier AI companies probably can’t leave the US
It’s plausible that, over the next few years, US-based frontier AI companies will become very…
Improving Internal Model Principle
Funded by the Advanced Research + Invention Agency (ARIA) through project code MSAI-SE01-P005This post was…
A Positive Case for Faithfulness: LLM Self-Explanations Help Predict Model Behavior
This is a summary of our new paper.TL;DR: Existing faithfulness metrics are not suitable for…
How Robust Is Monitoring Against Secret Loyalties?
If monitoring is robust, secret loyalties may be very hard to act on—at least for…