Why language models make things up
Understand the causes of hallucination and build habits that catch it before it costs you
A taste of a lesson
It gave me a citation with authors, journal and year. How can it be fake if it is so detailed?
Detail is exactly what a language model is good at producing. It has seen thousands of real citations, so it knows the pattern: plausible author names, a journal that fits the field, a sensible year. When it lacks a real source, it can generate one that fits the pattern perfectly. Formatting tells you nothing about existence. So treat every citation as unverified until you find it in a library database or the journal itself. Try it now: search for the exact title. What do you find?
Written by the teacher as an example. In your lesson the tutor answers your own questions, and like any AI it can be wrong.
What you will be able to do
- Explain the main causes of hallucination in plain or technical terms
- Recognise common forms such as invented citations, fake quotes and unfaithful summaries
- Apply grounding and quoting techniques that reduce hallucination
- Decide how much verification a claim needs based on its specificity and stakes
Lesson plan
- 1 Fluent is not the same as true Understand why a model can produce confident false statements. Start
- 2 The many forms of made up content Recognise the typical shapes hallucinations take. Start
- 3 Why models guess instead of abstaining See how training and evaluation pressures encourage answering over admitting uncertainty. Start
- 4 Grounding: give it the facts Use provided sources and quoting to reduce unsupported claims. Start
- 5 Verification habits that scale Build a proportionate checking routine for real work. Start
Try asking
About this tutor
For anyone who relies on AI output, from casual users to professionals. You learn why fluent, confident falsehoods happen: the next token objective, gaps and conflicts in training data, the pressure to always produce an answer, and the absence of a built in fact check. You see the common forms, including invented citations, wrong numbers, fake quotes and plausible but false reasoning. Then you learn what reduces the risk, such as grounding in provided sources, retrieval and search tools, asking for quotes and uncertainty, and independent verification, and what does not reliably help. Lessons use real patterns and small experiments you can run yourself.
Reviews
4.4
5 ratingsSample
- Helena R.Sample
The niche topic experiment was humbling: four of five sources were fake. The 'specific, rare, consequential' checklist is now pinned above my desk.
- Pedro A.Sample
I work in a law office. The real sanctions cases and the quoting technique were exactly what our team needed.
- Greta N.Sample
Very good, if slightly long on causes before getting to fixes. The grounding lesson changed how I summarise reports.
- Jamal T.Sample
As a developer I liked the explanation of why evaluations reward guessing. Clear on what helps and what is just a placebo prompt.
- Zoe C.Sample
Useful but I hoped for actual hallucination rates by model. The tutor explained why it would not give single numbers, which is fair, just not what I wanted.
About the teacher
I explain how language models really work, from tokens to attention, without hand waving
9 tutors 525 lessons taught Sample
I like taking the mystery out of language models. I teach what happens between typing a question and reading an answer: tokens, context windows, embeddings, attention, training and fine tuning. I start every topic with a picture or a small worked example and only add maths when it earns its place. My work background is in software and teaching workshops...
See Bastian's profile and tutorsMore like this
Other tutors on the same or nearby topics.