Yifan Wu(@yifannnwu):Been thinking about this for a while, as tasks go to more and more turns and longer horizons, PPO is much more elegant for giving dense, per-turn reward. And here we go.

Yifan Wu

@yifannnwu

吴奕凡; AI Research Scientist @Meta | Ph.D. @penn @picslupenn @GRASPlab.

加入 November 2016

490 正在關注 1.4K 粉絲

Yifan Wu@yifannnwu

2026.06.17 02:36

Been thinking about this for a while, as tasks go to more and more turns and longer horizons, PPO is much more elegant for giving dense, per-turn reward. And here we go.

Sriraam@27upon2

2026.06.16 19:34

We’re going back to PPO everybody. Curious if there’s some research or experiments comparing the stability with GRPO style setups when you have compacted rollouts