Hallucination
In one line: When an AI confidently states something false. The biggest reliability issue with LLMs - Understanding hallucinations helps you use AI more safely.
What is Hallucination?
Hallucination is when an LLM produces output that is confidently stated but factually wrong. The model is not 'lying' in any intentional sense - it is doing what it always does (predicting statistically likely next tokens) but the plausible thing happens to be false. It is the most important reliability limitation of current AI systems.
Types of hallucination
- Citation hallucination - Inventing academic papers, journal articles, authors, or DOIs that do not exist. This is arguably the most dangerous type because the output looks authoritative and is hard to spot without checking the source.
- Factual confabulation - Stating wrong dates, statistics, biographical details, historical events, or scientific facts with the same confident tone as correct information.
- Code hallucination - Inventing function names, API methods, or library arguments that do not exist. Common with niche libraries or recent API changes after the model's knowledge cutoff.
- Quote hallucination - Attributing a plausible-sounding quote to a real person who never said it. Particularly problematic in journalism and academic contexts.
Why hallucination happens
LLMs predict statistically likely next tokens based on patterns in training data. They do not have an internal verified-facts database to cross-check against. When training data is sparse on a specific topic, the model fills the gap with a plausible-sounding fabrication rather than saying 'I don't know.' The model has no reliable way to distinguish between what it 'knows confidently' and what it is guessing - it just generates the most probable continuation.
Reasoning models (o3, DeepSeek R1) hallucinate less on structured logical and mathematical problems because their extended thinking process provides more opportunity for self-correction. But they still hallucinate on factual recall tasks.
- Use Perplexity or web-grounded models for factual queries that require up-to-date cited sources.
- Use RAG to ground answers in your own verified documents rather than relying on the model's training memory.
- Ask the model explicitly: 'Are you confident in this? What might you be wrong about?' Models trained with alignment techniques will often flag uncertainty.
- Always verify citations, statistics, and key claims independently before publishing or acting on them. The FAQ on using AI for research shows a claim-source-date table format that makes unsupported claims easy to spot.
- Use prompt templates that instruct the model to cite its sources and flag when it is unsure.
Hallucination rates by model
| Model generation | Relative hallucination rate | Notes |
|---|---|---|
| GPT-3 era (2020-2022) | High baseline | Frequent fabrication on factual recall |
| GPT-4 / Claude 2 (2023) | Meaningfully lower | Better calibration, still problematic for citations |
| GPT-4o / Claude Sonnet 4 (2024-25) | Lower still | Measurable improvement; RLHF and Constitutional AI help |
| Reasoning models (o3, DeepSeek R1) (2025) | Lowest on logic tasks | Extended thinking reduces math/code errors; factual recall still imperfect |
The problem is not solved. Treat AI output like a first draft from a knowledgeable but sometimes unreliable expert - valuable input that needs human verification for anything consequential. See also: alignment, RAG, knowledge cutoff.
Hallucination example
If you are using AskAI.free, a practical way to understand hallucination is to ask a model to explain it, then ask for a concrete example in your own workflow. For example: "Explain hallucination for someone using AI to write, code, research, or create images."
This turns the term from a dictionary definition into a decision-making tool: you can see when it affects prompt quality, model choice, output reliability, privacy, cost, or how much context the AI can use.
Why Hallucination matters
Hallucination matters because it changes how you choose, prompt, compare or trust AI systems. If you understand this term, you can ask better questions, spot weak answers faster and choose the right model or tool for the job.
A common mistake is treating hallucination as isolated jargon. It usually connects to nearby ideas like Inference and Jailbreak, so check those next if you want the full picture.
Common mistake with Hallucination
The most common mistake is using the term as a label without changing behavior. When hallucination comes up, ask what action should change: the prompt, the model, the input length, the evidence you request, or the way you verify the answer.
See it in action - Ask any AI about hallucination on AskAI.free.
Try it free →