Alexa Pan

Message

845

Latent reasoning architectures would likely undermine CoT, our strongest oversight tool

by Lukas Finnveden, Alexa Pan, Alek Westover, Girish Gupta, frisby, and ryan_greenblatt

Edited 2026-09-29; see note at the start of the conclusion. Summary: Currently, “chain of thought” (CoT) is our most valuable tool for understanding the reasoning and cognition of AI systems. However, some architectures would enable AI models to reason much more extensively in latent states rather than in text CoT....

Sep 23•95

An operationalization of opaque serial depth

by ryan_greenblatt, frisby, Alek Westover, Lukas Finnveden, Alexa Pan, and Julian Stastny

Currently, chain-of-thought (CoT) is a valuable tool for overseeing AI models. However, some architectural shifts could significantly reduce CoT monitorability. We have recently proposed that AI companies should transparently share information about the degree to which their architectures may allow for latent reasoning and communication. To assist with this proposal,...

Sep 10•73

Sonnet 4.5's eval gaming seriously undermines alignment evals, and this seems caused by training on alignment evals

According to the Sonnet 4.5 system card, Sonnet 4.5 is much more likely than Sonnet 4 to mention in its chain-of-thought that it thinks it is being evaluated; this seems to meaningfully cause it to appear to behave better in alignment evaluations. So, Sonnet 4.5’s behavioral improvements in these evaluations...

Oct 30, 2025•144