From Raw Model to Helpful Assistant

A base model continues text instead of answering it.

That one rule has an odd result: give a base model a request, and it often replies with more requests. Type "Write a Python function that reverses a string." The model might add two more lines: "Write a Python function that checks for a palindrome. Write a Python function that counts vowels." The function you asked for never appears, yet the model is working exactly as it was trained.

A base model has finished only the first stage of training, which teaches it to predict the next piece of text