Fetching the paper…

Model Tampering Attacks Enable More Rigorous Evaluations of LLM Capabilities · Around