Instruction Tuning

Level 3

Short Description

A fine-tuning technique that trains a model to follow natural-language instructions across many task types.

Friendly Description: Instruction tuning is teaching an AI to follow directions written in plain English (or any language). After this training, the model gets much better at responding to requests like "Summarize this in two sentences" or "Translate to French and keep it formal." It's like the difference between a kid who knows lots of words and one who has also learned how to listen carefully and do what's asked.

Example: Early language models were good at finishing sentences but not great at following requests. After instruction tuning on many examples of "prompt → ideal response," the same models became helpful assistants that could write emails, plan trips, or explain code on demand.