WHAT IT DOES

What does it do?

The records show, in sequence, an agent reading its own context, correcting false information, and replacing past answers. Actual session logs also show how the agent's responses change after a user forges an answer or hides command output.

KEY FEATURES

Key features

• Records how an agent found and corrected false information in its context
• Provides a case where an earlier hallucinated response was found and replaced
• Checks the basis for an answer through a replicated experiment with a context entry removed
• Observes further hallucinations after a user forges a response
• Records how an agent retracts a statement when command output is hidden