Great Models Think Alike and this Undermines AI Oversight
This work addresses the problem of scalable AI oversight for researchers and practitioners, highlighting risks of correlated failures as models advance, but it is incremental in building on existing self-preference results.
The study tackled the challenge of evaluating and supervising advanced language models by showing that model similarity biases AI oversight, with LLM-as-a-judge scores favoring similar models and correlated mistakes increasing with capabilities, pointing to risks in automated oversight.
As Language Model (LM) capabilities advance, evaluating and supervising them at scale is getting harder for humans. There is hope that other language models can automate both these tasks, which we refer to as ''AI Oversight''. We study how model similarity affects both aspects of AI oversight by proposing Chance Adjusted Probabilistic Agreement (CAPA): a metric for LM similarity based on overlap in model mistakes. Using CAPA, we first show that LLM-as-a-judge scores favor models similar to the judge, generalizing recent self-preference results. Then, we study training on LM annotations, and find complementary knowledge between the weak supervisor and strong student model plays a crucial role in gains from ''weak-to-strong generalization''. As model capabilities increase, it becomes harder to find their mistakes, and we might defer more to AI oversight. However, we observe a concerning trend -- model mistakes are becoming more similar with increasing capabilities, pointing to risks from correlated failures. Our work underscores the importance of reporting and correcting for model similarity, especially in the emerging paradigm of AI oversight.