LLM-as-Judge
An evaluation method that uses a large language model as an automatic judge to classify or score model outputs for safety. It observes only the final text and cannot determine the state of internal safety mechanisms.
Timeline (last 90 days)
Minimal supervised fine-tuning can achieve LLM-judge scores comparable to multi-stage reinforcement learning pipelines.
The reward-free filtering uses a consensus mechanism (sampling takeovers multiple times and keeping segments with consistent outcomes) or LLM judging (a large model judges whether the takeover is reasonable) to filter out low-quality trajectories, and the entire process does not rely on task success signals.
All Facts (2)
Minimal supervised fine-tuning can achieve LLM-judge scores comparable to multi-stage reinforcement learning pipelines.
50%UnverifiedThe reward-free filtering uses a consensus mechanism (sampling takeovers multiple times and keeping segments with consistent outcomes) or LLM judging (a large model judges whether the takeover is reasonable) to filter out low-quality trajectories, and the entire process does not rely on task success signals.
50%