An AI hallucination is a confident, fluent-sounding response from a language model that is factually wrong or entirely made up, rather than an honest admission that it doesn't know.
A hallucination happens when an AI model generates information that sounds plausible and is stated with confidence, but is factually incorrect, fabricated, or unsupported by any real source. This can range from a wrong date or statistic to an entirely invented citation, quote, or event.
Language models generate text by predicting the most statistically likely next word, not by looking up verified facts in a database. When a model doesn't have reliable information about a topic, it will still produce fluent-sounding text, because generating plausible language is what it was trained to do, whether or not that language is true.
Retrieval-augmented generation reduces hallucination by giving a model real source documents to reference instead of relying purely on memorized patterns. Fine-tuning a model to say "I don't know" when it's uncertain, and having systems cite sources so claims can be checked, are other common mitigations. No current technique eliminates hallucination entirely.
Hallucination is one of the central trust problems in deploying AI systems, especially in fields like medicine, law, and journalism where a confidently wrong answer can cause real harm. It's also why most serious AI products now show sources or confidence signals rather than presenting model output as unquestionable fact.
Researchers have developed a way to make AI models better at admitting when they don't know something. This could make AI assistants more reliable in everyday use. The method works without needing to see the model's internal workings, making it useful for commercial AI services.
Researchers discovered that providing one confirmed fact in a multi-step reasoning chain increases LLMs' likelihood of confidently producing wrong answers. This phenomenon, called anchored confabulation, challenges assumptions about how models handle partial evidence.
Researchers introduce a dynamic data curation pipeline to better evaluate how LVLMs handle conflicts between visual and textual evidence. The study underscores the importance of models admitting when they lack knowledge.
SymptomWise introduces a hybrid framework that separates language understanding from diagnostic reasoning to eliminate hallucinations in AI symptom analysis. By combining expert-curated knowledge with deterministic inference, the system ensures traceable and consistent outputs in safety-critical settings.
Researchers propose a novel framework treating LLM hallucinations as output-boundary misclassification errors, introducing a composite abstention architecture. This system combines instruction-based refusal with a structural gate that blocks unsupported claims based on a calculated support deficit score.