Machine Learning

Instruction Tuning

A fine-tuning approach where a model is trained on a dataset of instruction-response pairs, teaching it to follow human instructions accurately. This transforms a text-completion model into a helpful assistant.

Why It Matters

Instruction tuning is what transforms a raw language model into something like ChatGPT or Claude — a model that understands and follows user requests.

Example

Training a model on thousands of examples like 'Summarize this article: [article]' → [summary], teaching it to follow the instruction format.

Think of it like...

Like training a new employee not just to know information but to respond appropriately to requests — 'When someone asks for X, do Y.'

Related Terms