News for hackers. Stuff that matters.
Re:CS234: Reinforcement Learning Winter 2025 by whatshisface on Tuesday November 25, 2025 @10:34PM (#46053928)
RL is barely even a training method, its more of a dataset generation method.
Re:CS234: Reinforcement Learning Winter 2025 by theOGognf on Tuesday November 25, 2025 @11:40PM (#46054259)
I feel like both this comment and the parent comment highlight how RL has been going through a cycle of misunderstanding recently from another one of its popularity booms due to being used to train LLMs
Re:CS234: Reinforcement Learning Winter 2025 by mistercheph on Wednesday November 26, 2025 @01:55AM (#46054846)
care to correct the misunderstanding?
Re:CS234: Reinforcement Learning Winter 2025 by mountainriver on Wednesday November 26, 2025 @09:38AM (#46057770)
I mean DPO, PPO, and GRPO all use losses that are not what’s used with SFT for one.

They also force exploration as a part of the algorithm.

They can be used for synthetic data generation once the reward model is good enough.

Re:CS234: Reinforcement Learning Winter 2025 by phyalow on Wednesday November 26, 2025 @04:41AM (#46055801)
Its reductive, but also roughly correct.
Re:CS234: Reinforcement Learning Winter 2025 by singularity2001 on Wednesday November 26, 2025 @09:03AM (#46057493)
While collecting data according to policy is part of RL, 'reductive' is an understatement. It's like saying algebra is all about scalar products. Well yes, 1%