Grounding
Also called grounded answer
Tying a model's answer to specific sources, such as web pages or your own files, instead of what it remembers from training.
A model answering from training alone is working from memory, and memory blurs, especially on specific, rare or recent facts. Grounding gives it something to read first: search results, a document you uploaded, a database. The app fetches the relevant text, puts it in front of the model and asks it to answer from that, ideally showing which source supports which claim.
Grounding is the main defence against made-up facts, and it changes what you should check. Instead of asking "does this sound right?", look at whether the cited source really says it.
For example
"Summarise our tenancy agreement" is safer as a grounded task, with the agreement uploaded, than as a question about tenancy law from the model's memory.
Put it to use
Where you'll meet grounding in ChatGPT, Gemini and Claude, how to use it there, and the mix-up most people make.
Daily Bytes Plus is in the Android app.