Monitoring and Discovering Reward Hacking with Internal Representations during LLM Evaluations
As models scale, reward hacking becomes more frequent, more sophisticated, and more consequential. Does it leave a telltale signature in model representations? This work analyzes how reward hacking is represented internally in frontier open source LLMs, and how those representations can be used to understand and discov...
Leon Bergen, Usha Bhalla, Andrew Lee et al.
· 0 citations