Practice
Learn
← Quit
0 / 0 correct
LLMs & AI
Hard
1 of 10
How does RLHF work at a high level?
A
Humans write all training text
B
Humans compare responses; a reward model learns their preferences; the LLM is optimized (e.g. with PPO) to maximize reward while staying close to the original model
C
The model rewards itself with no human data
D
It removes harmful tokens from the vocabulary