TinkerThe exam hall: the model, a stack of cards per written answer as tall as its odds, and a grader’s desk.
Chapter 1 of 7
Base or assistant
Larkspur wants an assistant of its own. Pre-training gave it a base model that has read a mountain of text, but it doesn’t answer customers yet.
This chapter: What is the difference between a base model and an assistant?
About this chapter
You will learn
- Why a base model continues text instead of answering
- What supervised fine-tuning on examples changes
- What post-training adds after that
Words you’ll meet
- Base model
- A model trained only to guess the next words of documents. It continues text; it was never taught to answer.
- Post-training
- Everything done to a base model after pre-training to turn it into a helpful assistant.
- Supervised fine-tuning
- Training on examples people wrote: a prompt and the reply they want. The model learns to copy their style.
Step 1 of 4
A customer asks about refunds. A continues text: it learnt only to guess the next words.
Press Ask the base model.
The customer asks
“How long do standard customers have to ask for a refund?”
The toy policy is a choice among answers written for this site, not a language model. Its updates run live in your browser.