# PPO

A policy-gradient method that clips the update size so each step stays close to the previous policy.

## What it is used for

The workhorse for scheduling and control agents, and the optimizer in RLHF-style language model tuning.