More capable models are better at in-context scheming — Apollo Research
We evaluate models for in-context scheming using the suite of evals presented in our in-context scheming paper (released December 2024) with the most capable new models.| Apollo Research
