Method

PPO

Proximal policy optimization.

6 cases from 4 companies apply PPO. 2 were checked in an online A/B test.

Typical tasks: Label data with LLMs, Build an assistant over internal docs, Recommend from what the user just did, Optimize decisions. Companies: Rabota.ru, Netflix, SberDevices and 1 more. Often combined with: Reward model, RLHF, Supervised fine-tuning, DPO / preference optimization, Synthetic data generation (LLM) and 18 more.