When AI gets it wrong: hallucinations, bias and limits
Why chatbots invent facts and sources, why confident answers are not correct ones, how bias and knowledge cutoffs work, and how to check an answer fast.
A chatbot has no way to signal that it does not know something. It answers a rare question in the same steady, well organized voice it uses for an easy one, which is exactly why wrong answers get past careful people. Knowing where these systems fail, and what each kind of failure looks like on the page, is the difference between a tool that saves you time and one that quietly creates work.
Begin with why a model invents things at all, then the fabricated sources problem, the version that has embarrassed lawyers, students and journalists in public. Knowledge cutoffs explain a different kind of wrongness, the answer that was true two years ago and is stale now, and why simple arithmetic goes wrong shows a third, because fluent text is not calculation.
None of this means you cannot use AI for factual work. It means you check before you pass anything on, and a five minute verification routine is enough for most everyday answers.
Articles in this hub
- Why AI makes things up, and what a hallucination really isThe reason chatbots state false things fluently, which questions trigger it most, and the simple habits that catch a made up answer before you rely on it.
- Why AI sounds certain when it is wrongFluency is not confidence and confidence is not accuracy: what the model is really doing when it states something firmly, and the phrasing that reveals a guess.
- Why an AI agrees with you even when you are wrongSycophancy explained: why chatbots fold when you push back, how training created the habit, and how to ask in a way that gets disagreement when disagreement is correct.
- Made up sources: why AI invents books, cases and studiesWhy chatbots produce citations that look perfect but do not exist, which fields it happens in most, and a fast routine for verifying any reference you are given.
- Knowledge cutoffs: why your AI does not know last weekWhat a training cutoff date means, why chatbots get recent events wrong or guess, and how to tell whether an answer came from memory or from a live web search.
- Why AI gets simple math wrong and when to trust itWhy a chatbot can explain calculus yet fumble arithmetic, how tokenization breaks number handling, and when a code or calculator tool makes the answer reliable.
- AI bias: where it comes from and where it shows upHow bias enters an AI system through data, labels and objectives, the places it surfaces in real decisions, and the questions that expose it before it costs somebody something.
- How to fact check an AI answer in five minutesA repeatable routine for checking what a chatbot told you: which claims to check first, how to verify a citation, and the three questions that catch most invented answers.
- Prompt injection: when a web page gives your AI new ordersHow hidden instructions in a document, email or web page can hijack an AI assistant, why the problem is hard to fix, and what it means once an assistant can act on your behalf.