jain18ayush

Executive summary We extend Anthropic's emotions vector work and observe strong evidence that a model’s internal representations of certain emotions change when it takes misaligned actions in two scenarios: reward hacking in code generation and specification gaming in a sales environment. 1. We tested two approaches: a) measure drift in...

We introduce model hermeneutics: studying a closed-weight model (the author model) through the internals of an open-weight substitute (the reader model). We find: 1. Weak to strong readers can work. A 27B reader was able to match the performance of probing a 397B author. 2. Probes on open-weight reader models...