This is a curated English edition of a legacy Portuguese article originally published on 2023-09-15. It preserves the public argument and editorial intent while avoiding private details, operational state, or unsupported new claims.
Core idea
The article explains hallucination as a predictable failure mode of systems trained to produce plausible language rather than guaranteed truth.
The operational lesson is simple and stubborn: use retrieval, evidence, tests, and review when factual accuracy matters. Confidence is not provenance.
Editorial note
This translation was prepared during the governed migration of the legacy blog into the GitHub Pages hub. Technical references may age; when the topic involves law, public policy, infrastructure, or model behavior, treat the article as educational context and verify current sources before acting.
Portuguese source article: /blog/entendendo-alucinacoes-em-grandes-modelos-de-linguagem/.