CCDV-F: Claude Certified Developer — Foundations

By Anthropic · Question Mode

QuestionQ78

VerifiedEval, Testing, and Debugging

A Claude application that performed well in testing is now occasionally producing outputs that mention information not present in the input. The development team initially assumed the model was hallucinating, so they asked you to troubleshoot.

What would you do first?

Explanation

Reviewing production traces helps distinguish hallucination from missing context, prompt injection, or another failure mode, so the remedy can target the actual cause. Unsupported output alone does not establish which failure occurred. Anthropic documents hallucination and prompt injection as distinct problems with different mitigations. (platform.claude.com)

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!