The probability an LLM assigns to a response is treated as a reward, formally justified via an equivalence between next-token prediction and offline inverse RL, and then used to fine-tune the model itself.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Generalist Reward Models: Found Inside Large Language Models
The probability an LLM assigns to a response is treated as a reward, formally justified via an equivalence between next-token prediction and offline inverse RL, and then used to fine-tune the model itself.