Why chatbots invent facts, and how to stop them from fooling you
Tages-Anzeiger explains why ChatGPT and Gemini make up facts, studies and links, and which habits cut down on embarrassing errors.
The short version
- Chatbots invent facts, studies and links, a problem called hallucination.
- Models err most when they lack reliable information on a topic.
- Very specific requests or too much input can overwhelm a model.
- Per OpenAI, common scoring methods reward guessing over admitting gaps.
Why it matters
Anyone who uses a chatbot for school, client work or published writing can end up repeating an invented claim. Knowing why the errors happen tells you where to check first. The fixes are simple habits, not technical skills.
What the problem looks like
As Tages-Anzeiger reported in a practical explainer, AI chatbots such as ChatGPT and Gemini sometimes invent facts, studies and web links. The industry calls this hallucination. The made-up material often sounds as confident as the real thing.
Why it happens
The source names three causes. First, models hallucinate most when they have too little reliable information on a subject. Second, requests that are too specific, or too much pasted input, can overwhelm a language model.
Third, OpenAI says the way models are scored pushes them to guess instead of admitting they do not know. A model that guesses can score better than one that says "I'm not sure".
Background from our own newsletter helps here. A language model predicts the next small piece of text, called a token, over and over. No step in that loop checks whether a statement is true.
Habits that help
The source promises methods to prevent false claims, but we only have its summary. The advice below is ours, not the article's.
- Treat any date, figure, quote, or legal or medical point as unverified until you check a primary source.
- Paste in the text you want the model to work from, rather than asking it to recall facts.
- Ask it to say when it is unsure.
- Ask for sources, and open them yourself.
Be most careful with names, citations and numbers. Invented detail in those areas reads just as convincingly as real detail.
What to watch
- Will AI makers change how they score models so that admitting uncertainty is no longer penalized?
- How well do the article's prevention methods hold up on real work such as school or client tasks?
Try it
Next time you ask a chatbot for a study or link, open every source it gives you. If one does not exist, ask it to work only from text you paste in.
Sources and links
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by Tages-Anzeiger · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.