Jargon, explained twice.
Once in plain English, for anyone. Once properly, for people who want to know what is actually happening under the bonnet. No assumed knowledge, no waffle, and every term linked to the ones it depends on.
Open the AI glossarySee the graph
577 terms · 27 sections · free to reuse under CC BY-SA 4.0
Start anywhere
- Large Language ModelA very large neural network trained to predict the next word in a piece of text.
- TokensThe small chunks of text a model actually works with.
- TransformerThe particular design of neural network that nearly all modern language models use.
- HallucinationWhen a model states something false with complete confidence.
- Fine-tuningGiving a pre-trained model additional training on a narrower task, domain or style.
- Retrieval Augmented GenerationA technique where the system searches a document collection for relevant material and includes it in the prompt before the model answers.
How an entry works
The top of every page is the plain version. It avoids jargon, and where it cannot avoid a term it links to that term instead of assuming you know it. Underneath sits a control marked Technical view. Open it and you get the mechanism: the maths, the trade-offs, the things practitioners argue about. Close it and the page stays readable.
At the bottom of each entry is a short list of related terms with a line saying why each one is related, which is more useful than a bare list of links, and a small picture of how that term sits in the wider web of terms.
Where it comes from
Jargonautica started as one person's notes and got out of hand. It is written and maintained by Big Trouble Labs, published under a Creative Commons licence, and open to suggestions for terms that are missing. AI is the first glossary. It will not be the last.