An LLM is fundamentally predicting the next most likely word — it doesn't actually verify facts, so it can confidently give wrong answers or invent paper titles and legal clauses. The sparser the training data in a domain, the more common hallucinations are.
Engineering mitigations include RAG (grounding answers in retrieved real material), requiring cited sources, and more conservative prompting. But any critical AI-generated information should still be verified by a human — which is why Enclave's terms state plainly that AI content is for reference only.
Try it yourself in Enclave
Open it in your browser — no credit card, no install.