auto-regression. auto meaning self. regression meaning predict. predict next token given previous tokens: P(t_t | t_1, t_2, ..., t_{t-1}). LLM training is self-surpervised learning.
Overview
Objective of LLM is to predict the next word in a sequence of words. This is done by using a model that is trained on a large corpus of text.
Levels for dealing with LLMs:
- Prompt Engineering: do not change any model parameters, only change the input prompt
- Fine-tuning: change the model parameters to better fit the data. Update model params given task-specific data
- Model Engineering: change the model architecture to better fit the data