1
Pretraining
Learning to predict next words
Objective: Train the model to incrementally predict the next word in a very large corpus of text.
CLICK TO EXPLORE
→
2
Finetuning & Alignment
Learning to perform tasks and to be helpful and not harmful
Objective: Train the model to follow instructions (e.g., answer questions, summarize, write code, translate) and optimize it to be maximally helpful and minimally harmful.
CLICK TO EXPLORE