Self-Critique
A mechanism in which an agent examines intermediate results during reasoning and backtracks to correct errors once found. It is often combined with chain-of-thought to form a closed loop, improving reliability on multi-step tasks.
Timeline (last 90 days)
The approach of using a model itself to assist in safety judgments is referred to in the industry as LLM-as-a-Judge or the model Self-Critique mechanism.
The model self-critique path typically needs to be used in combination with other defense layers such as sandbox isolation and post-output monitoring, rather than being relied upon alone.
OpenAI's research shows that having models engage in self-critique can significantly reduce reasoning error rates.
All Facts (3)
The approach of using a model itself to assist in safety judgments is referred to in the industry as LLM-as-a-Judge or the model Self-Critique mechanism.
50%UnverifiedThe model self-critique path typically needs to be used in combination with other defense layers such as sandbox isolation and post-output monitoring, rather than being relied upon alone.
50%UnverifiedOpenAI's research shows that having models engage in self-critique can significantly reduce reasoning error rates.
50%