Игорь Котенков RLHF Intro: from Zero to Aligned Intelligent Systems

A story about Text Summarization What the Alignment is, and what s the problem How RLHF works Data setup, and why we d like to follow instructions Reward Modeling and PPO Why RLHF works (and when it doesn t) ChatGPT improvements What s next and what to expect Data Fest 2023: Трек Instruct Models : Наши Telegram: t. me, datafest Вконтакте: