Inconsistencies, Bias, and Hallucination-Reduction Techniques

Three Different Kinds of "Unexpected"

Not every problem in an output is a Hallucination. Recognizing which kind of "unexpected" it is determines the correct response:

• **Hallucination**: a fabricated fact, citation, or quote that doesn't exist.
• Inconsistency: an internal contradiction -- a total that doesn't match its line items, a recommendation that conflicts with an earlier statement in the same output.
• Bias: skewed framing, unrepresentative examples, or assumptions about people.

Techniques That Reduce Hallucination Risk

From Anthropic's reduce-hallucinations guidance, four techniques lower (but do not eliminate) the risk:

1. Ground the request in provided source material and ask Claude to answer only from it.
2. Allow an "I don't know" exit so the model isn't pushed to guess when it lacks the answer.
3. Ask for citations/quotes that can be traced back to the source.
4. Cross-check specific claims against an authoritative source before relying on them.

Reduce, Not Eliminate

Even with grounding or retrieval-augmented generation (RAG) in place, the model can still misread or ignore the provided context and produce an unsupported claim. Grounding lowers the odds of fabrication; it does not guarantee correctness. The cross-check step remains necessary even after grounding.

Common exam traps

• "Asking Claude how confident it is tells you if it's right." Self-reported confidence is not a reliable accuracy signal -- a hallucinated answer can be stated just as confidently as a correct one.
• Assuming grounding/RAG eliminates Hallucination entirely. It *reduces* the risk, but the model can still misread or ignore the supplied context.