Logits as a new monitor for evaluation awareness
Builds a logit-based monitor that estimates the probability of eval-aware sentences in a chain of thought, outperforming LLM-judge monitoring with 10-100x fewer rollouts on Kimi K2.5 and Qwen 3 32B.
Understanding and steering neural networks. goodfire.com
Builds a logit-based monitor that estimates the probability of eval-aware sentences in a chain of thought, outperforming LLM-judge monitoring with 10-100x fewer rollouts on Kimi K2.5 and Qwen 3 32B.