Anastasiya Ovchinnikova, data scientist, Rabota.ru (works on recommendations)
The talk describes how the team applied RLHF to the task of summarizing job vacancies for the Rabota.ru site feed: why summaries are needed, why an abstractive approach and the T5 model were chosen, how the dataset for the reward model was collected, and how the PPO pipeline is structured. Then it presents the results, a preliminary A/B test, and the limitations.
Continue reading — it's free
Google or a one-time email code. No password.