Hallucination (LLM)

A hallucination is an output from a language model that sounds plausible and confident but is factually incorrect, fabricated, or unsupported by the information the model was given. The danger is precisely that hallucinations are fluent: the grammar is perfect and the tone is authoritative, so a wrong answer reads exactly like a right one. This is the single most important failure mode to understand before trusting an LLM with technical work like mining diagnostics or firmware questions.

Why it happens

A language model is trained to predict the most likely next token, not to retrieve verified facts. When the training data is sparse, ambiguous, or absent for a query, the model still produces its best statistical guess rather than saying "I don't know." The result can be invented citations, fake part numbers, or confidently stated specifications that do not exist. Hallucination is intrinsic to how these systems work, not a bug that a single patch removes.

Reducing it

The most effective mitigation is grounding the model in verified data through retrieval-augmented generation, where the system fetches authoritative documents and the model answers from them rather than from memory. Lowering temperature, asking for sources, and validating output against trusted references all help, but none make hallucination zero.

For a sovereign operator, this is why D-Central treats a model as a drafting tool, never a source of truth: every factual claim about mining hardware is cross-checked against verified documentation. An AI agent that can look up real data will hallucinate far less than one answering from weights alone.