LLM overview in 2024

1. Why SFT is before RLHF
   • RLHF need basic capability from the model.
2. 2 steps in RLHF
   If only learns from HF, then you knows answer to a specific question, but NOT to general questions.
   • Learn Human preference by training a Reward Model to give rewards for answers
   • Adjust network output based on the rewards
3. Alignment
   SFT + RLHF = Alignment (to human)