Model Self-Explanation
A method that has large language models explain their own reasoning process or basis for decisions in natural language. The reliability of such explanations is contested, including concerns about rationalization.
Timeline (last 90 days)
Research shows that large language models may generate explanations that seem plausible but do not match the actual computation process—the phenomenon known as "rationalization."
When tracing the source of agent failures, relying on model self-explanations requires caution and is best cross-validated with other evidence (such as traces and tool outputs).
All Facts (2)
Research shows that large language models may generate explanations that seem plausible but do not match the actual computation process—the phenomenon known as "rationalization."
50%UnverifiedWhen tracing the source of agent failures, relying on model self-explanations requires caution and is best cross-validated with other evidence (such as traces and tool outputs).
50%