Following up on our previous work on verbalized eval awareness:we are sharing a post investigating the emergence of metagaming reasoning in a frontier training run.Metagaming is a more general, and in our experience a more useful concept, than evaluation awareness.It arises in frontier training runs and does not require training on honeypot environments.Verbalization of metagaming can go down over the course of training.We also share some quantitative analyses, qualitative examples, and upcoming work.Discuss Read More
Metagaming matters for training, evaluation, and oversight
Following up on our previous work on verbalized eval awareness:we are sharing a post investigating the emergence of metagaming reasoning in a frontier training run.Metagaming is a more general, and in our experience a more useful concept, than evaluation awareness.It arises in frontier training runs and does not require training on honeypot environments.Verbalization of metagaming can go down over the course of training.We also share some quantitative analyses, qualitative examples, and upcoming work.Discuss Read More
Related Posts
AntiPaSTO: Self-Supervised Value Steering for Debugging Alignment
Published on January 13, 2026 12:55 PM GMTAntiPaSTO: Self-Supervised Value Steering for Debugging AlignmentDemo: Same…
Inoculation prompting: Instructing models to misbehave at train-time can improve run-time behavior
Published on October 8, 2025 10:02 PM GMTThis is a link post for two papers…
I am a rationalist
Published on November 24, 2025 6:39 AM GMTI am a relatively central instance of what…