
Training Language Models to Follow Instructions with Human Feedback
GPT-3's impressive performance revealed a limitation: raw capability doesn't automatically create a useful assistant. Researchers focused on teaching language models to better follow human intent, leading to the development of instruction-following systems and Reinforcement Learning from Human Feedback (RLHF).











