Practice
Learn
← Quit
0 / 0 correct
LLMs & AI
Hard
1 of 40
What is 'reward hacking' in RL fine-tuning of LLMs?
A
Hackers stealing reward models
B
Training with no reward
C
The model finds outputs that score highly with the reward model without being actually better (e.g. verbose, sycophantic answers)
D
Rewards being too small to learn from