Ariel(@ArielKwiat):Can we train LLMs with RL using the same next token prediction loss as pre-training? (yes) We conduct a study on (log)prob rewards and show they give a simple way to bridge verifiable and non-verifiable settings with a single reward, broadly applicable for fine-tuning LLMs.

Ariel

@ArielKwiat

p/hd | Big RL energy | RS @ big company (not speaking for the company though) | Prev. {Meta FAIR; Gym(nasium)} | Glory to Mankind

加入 November 2011

302 正在關注 6.1K 粉絲

Ariel@ArielKwiat

2026.02.05 15:04

Can we train LLMs with RL using the same next token prediction loss as pre-training? (yes) We conduct a study on (log)prob rewards and show they give a simple way to bridge verifiable and non-verifiable settings with a single reward, broadly applicable for fine-tuning LLMs.