Your AI SaaS business, ready by this week!
Try demoYour AI SaaS business, ready by this week!
Try demoAn AI hallucination is a confident answer from a model that is false or made up, such as an invented fact, quote, link or source.
Language models generate the most likely next words, not verified facts. When the training data is thin or the question is outside what the model knows, it can produce a fluent answer that is simply wrong.
Hallucinations are most common with specific details: numbers, dates, names, citations and URLs. Newer and larger models hallucinate less, but none are immune.
The usual fixes are grounding and instructions: give the model the source text (RAG or web search), tell it to answer only from that text, and show users where the answer came from.
Assistants answer from their knowledge base, chat can search the web for current facts, and you can instruct assistants and chatbots to say when they do not know.
Discover more terms connected to this one
RAG (retrieval-augmented generation) is a technique where relevant passages are first retrieved from a knowledge base and then given to an AI model with the question, so the answer is grounded in that content.
Grounding is giving a model trusted source material, such as documents or search results, and asking it to answer from that material to reduce made-up answers.
AI web search lets a chat model search the internet during a conversation and answer from current pages, with sources, instead of only from its training data.