Hallucination
Also known as: AI hallucination, confabulation.
In plain English
LLMs generate the most-statistically-likely next token, not the most-truthful one. When training data is incomplete or the model's internal pattern matches a plausible-but-wrong answer, it asserts confidently. Hallucinations cluster in two areas: open-ended assertion ('what was X's revenue in 2022?') and creative gap-filling ('explain why...'). Mitigations include multi-model consensus (different models hallucinate different things, so agreement is reliable), citation requirements, and post-generation hallucination detectors.
Example
A lawyer asks an LLM 'cite the leading US Supreme Court case on workplace privacy'. The model invents a plausible-sounding case name + citation that doesn't exist. A multi-model check would catch this — only one of the three models would 'know' the case, and the disagreement is the signal.
Hallucination in Vincony
Vincony's Hallucination Detector — now the AI-output tab inside the Verify super-tool — cross-references AI claims against the sources actually retrieved. The Consensus Engine queries 3 models in parallel and surfaces disagreement as the signal to investigate further.
Open the Hallucination Detector tab in VerifyTry it — 750+ distinct models across 80+ providers on one account
Vincony bundles GPT-5, Claude, Gemini, Perplexity Sonar Pro, DeepSeek, Mistral, and 750+ other models on one $0/month account.