Arun Jose

It would be useful if we had the ability to modify a model’s beliefs. For example, this could facilitate honeypots and better monitoring[1], help us do better science on current models[2], and augment certain forms of alignment training[3]. Currently, the state-of-the-art method for belief editing is synthetic document finetuning (SDF)....

The CoTControl eval asks reasoning models to follow formatting constraints in their chain-of-thought (e.g. write in all lowercase, avoid a specific word) while solving questions. Models seem to mostly be pretty bad at this: recent models score between 0-30% with the exception of Mythos Preview[1]. OpenAI and Anthropic have used...

We’d like to use powerful AIs to answer questions that may take a long time to resolve. But if a model only cares about performing well in ways that are verifiable shortly after answering (e.g., a myopic fitness seeker), it may be difficult to get useful work from it on...

Many technical AI safety plans involve building automated alignment researchers to improve our ability to solve the alignment problem. Safety plans from AI labs revolve around this as a first line of defence (e.g. OpenAI, DeepMind, Anthropic); research directions outside labs also often hope for greatly increased acceleration from AI...

TL;DR: Simple inoculation prompts that prevent misalignment generalization in toy setups don't scale to more realistic reward hacking. When I fine-tuned models on realistic reward hacks, only prompts close to the dataset generation prompt were sufficient to prevent misalignment. This seems like a specification problem: the model needs enough information...

TL;DR: Models trained with outcome-based RL sometimes have reasoning traces that look very weird. In this paper, I evaluate 14 models and find that many of them often generate pretty illegible CoTs. I show that models seem to find this illegible text useful, with a model’s accuracy dropping heavily when...

TL;DR: I made a dataset of realistic harmless reward hacks and fine-tuned GPT-4.1 on it. The resulting models don't show emergent misalignment on the standard evals, but they do alignment fake (unlike models trained on toy reward hacks), seem more competently misaligned, are highly evaluation-aware, and the effects persist when...