Your LLM doesn't admit when it doesn't know the answer. How do you fix it?
LLM & AILevel: IntermediateTroubleshooting Scenarios
Your LLM doesn't admit when it doesn't know the answer. How do you fix it?
Hints
- Identify the stage: pretraining, adaptation, generation or evaluation.
- Name the inputs, constraints and a case where the approach fails.
Answer
- System prompt calibration: explicitly instruct "If you are not certain, say 'I don't know' instead of guessing"
- Include confidence instruction: "Rate your confidence 1–5. If below 3, decline to answer"
- RAG with retrieval threshold: if no context was retrieved above a similarity threshold, route to a fallback response
- Fine-tuning with abstention examples: train on examples where the correct output is "I don't know"
- Evaluator chain: a second LLM call checks whether the answer is grounded in available context
Source and licence: JustInCache & contributors — Awesome AI Interviews · Apache-2.0