Skip to content
ReviewRamble

The weekly brief

AI, reviewed, in your inbox.

  • New reviews of tools, models, MCP servers and repos, with the verdict up front
  • Re-checked prices and ratings when something changes
  • New analysis pieces as they are published

Models, agents, MCP servers and the tools actually worth your time.

One email a week. Unsubscribe with the link in any brief. How we use your email

Hallucination (AI)

A hallucination is output from an AI model that sounds fluent and confident but is false, unsupported by its sources or made up, such as an invented citation, quote, statistic or API function.

Also called: confabulation, invented answers, AI hallucination

By Divya Sharma · Checked Nov 15, 2026 · How we review

Language models are trained to produce likely text, not to check facts. When a model does not know an answer, the most likely continuation can still be a confident, specific and wrong one. That is a hallucination. Some researchers prefer "confabulation", because the model is not perceiving anything; it is filling a gap.

Common forms

A model may invent a paper or court case, attribute a quote to the wrong person, give a wrong date, or call a library function that does not exist. In summarisation, it may add a detail that is not in the document it was given. That second kind matters most when you build on top of your own data, because the user assumes the answer came from the source.

Why it happens

A 2025 paper by Kalai and colleagues argues that training and evaluation reward guessing: a test that gives no credit for "I don't know" pushes a model to guess. Gaps in training data, a knowledge cutoff, and prompts that assume a false premise all raise the risk.

Reducing it

No current method removes hallucination completely. What helps: retrieval-augmented generation so the model answers from supplied text, asking for citations you can check, grounding answers in tool results, letting the model say it is unsure, and guardrails that check outputs before users see them. Measure the rate on your own tasks with LLM evaluation rather than trusting a general claim, because it varies by model, topic and prompt.

Related terms

Retrieval-augmented generation
Retrieval-augmented generation (RAG) is a pattern where a system first searches a collection of documents for passages relevant to a question, then gives those passages to a language model so its answer draws on them.
Grounding
Grounding is tying a language model's answer to specific, checkable sources, such as search results, documents or a database, so that its claims rest on supplied evidence rather than only on training data.
Guardrails (AI)
Guardrails are runtime checks placed around a language model that inspect inputs and outputs and block, rewrite or redirect them, so an application stays on topic, avoids harmful content and follows rules the model alone would not reliably keep.
LLM evaluation
LLM evaluation, often called evals, is measuring how well a model or AI application performs a task by running it on a set of test cases and scoring the outputs against defined criteria, by code, by humans or by another model.
Knowledge cutoff
A knowledge cutoff is the date after which a model has little or no information, because its training data was collected up to around then. Events after it are unknown unless supplied in the prompt.

Sources