Top postsTop post
(see full author list at the end) About a year ago, METR showed that the length of tasks frontier models can reliably complete doubles every few months. A related safety-relevant question is this: what length of tasks can models complete without any chain of thought (CoT)? We investigate in our new paper. If models can do extensive reasoning without outputting any CoT, it would have implications for safety. Developers and deployment-time monitors couldn’t easily understand models’ motivations and catch dangerous planning. Models that reason substantially without a CoT might also drift further from human patterns of thought, since their reasoning is no longer constrained by text in the pretraining prior. As a result, they would be harder to understand and might be more likely to scheme. Extending Ryan Greenblatt’s research, we investigate this by measuring models' ability to complete tasks without any CoT on a suite of 43 benchmarks spanning different domains. We compare AI reasoning ability to humans using the estimated 50% time horizon (TH)---the typical time taken for a human to perform a task that the LLM performs with 50% success rate. We find that frontier models like GPT-5.5 answer questions that take humans roughly three minutes with 50% reliability, and this TH has doubled approximately every year since 2019. Figure 1: Our no-CoT THs (green) compared to METR’s with-CoT THs (purple). Until the release of GPT-4, with- and without-CoT THs increased at a similar rate. Since GPT-4 with-CoT THs have grown at roughly twice the rate of no-CoT THs. We suggest that AI companies start to track no-CoT THs explicitly to find a lower bound on how much reasoning a model could do without revealing it to a CoT monitor or human. Our test suite does not require substantial inference compute so these evaluations are cheap to run. Methods We evaluated 14 frontier models from GPT-2 (2019) through GPT-5.5 (2026) on 43 benchmarks covering math, coding, knowledge, agentic to
I concluded my MARS 4.0 project titled 'Goal Crystallisation' with Anaïs Berkes and Lukas Gebhard under the mentorship of @Cameron Tice and @Jason Brown. We wanted to find out how important a threat scheming was. In particular, we wanted to find out whether a perfectly alignment faking agent could preserve...
This is a linkpost for https://arxiv.org/abs/2606.31591. Work done with Patrick Leask and Lev McKinney during the Astra Fellowship. TL;DR: Optimiser choice strongly influences emergent misalignment, while model size and family seem to barely matter. Optimisers that concentrate the LoRA update into fewer directions degrade alignment more, but regularising towards a...
(see full author list at the end) About a year ago, METR showed that the length of tasks frontier models can reliably complete doubles every few months. A related safety-relevant question is this: what length of tasks can models complete without any chain of thought (CoT)? We investigate in our...
Summary Safe deployment of an AI system requires that we can make confident claims about its behaviour on out-of-distribution deployment inputs on the basis of only pre-deployment evaluations. One approach to making such claims is to take a cognitive perspective, in which we interpret the AIs behaviour in terms of...
Summary Geodesic is going to use prediction markets to select their projects for MARS 4.0 and we need your help to make the markets run efficiently! Please read through the proposals, and then trade on the markets for the proposals you think might succeed or fail. We intend to choose...
Preamble This post is my attempt to try and organise some thinking about AI alignment in a way that will act as a partial-overview to the core ideas and approaches. It is mostly a review of existing ideas arranged with some light opinions thrown in. I do not expect the...