Practice
Learn
← Quit
0 / 0 correct
LLMs & AI
Hard
1 of 333
Why can reasoning models trained with RL on verifiable rewards (math, code) improve more reliably?
A
Correctness can be checked automatically (tests pass, answer matches), giving a reward signal that is hard to game
B
They avoid tokens entirely
C
They use larger vocabularies
D
They don't need any training data