Learn from the demonstrated answer

A journey through supervised fine-tuning with Botie.

Adapting and Evaluating Modern Models · A journey with Botie

First visit? Learn the workshop controls

Botie's notebook

Why Does a Text Predictor Need Examples of Following Instructions?

Part 1 of 3

A model trained to continue text has seen questions, answers, stories, arguments, and many other patterns. Given a request, it may continue the wording without doing what the person wanted. Predicting plausible text and providing a useful response are related tasks, but they are not identical.

Sources for this history

Ouyang and colleagues (2022), Training Language Models to Follow Instructions with Human Feedback

← Previous lesson: Diffusion Models Read the full lesson and derivations