Never Trust An LLM

LLMs hallucinate frequently and in multiple ways, including factual errors, fabricated entities, and contextual inconsistencies. Drawing on academic taxonomy papers and an OpenAI study, the core reason is that training and evaluation benchmarks reward guessing over admitting uncertainty. Practical mitigation strategies include always providing relevant context directly in the prompt and using the 'use your search tool' instruction to force web retrieval rather than relying on potentially stale or compressed training data. For critical decisions involving health or legal matters, always verify LLM outputs against primary sources yourself.