RLHF Challenges
Nathan discusses the complexities of RLHF, highlighting the struggle to balance helpfulness without verbosity in model responses. He points out that the introduction of new features, like toggling verbosity, requires sophisticated training and dataset management. Additionally, he notes OpenAI's acknowledgment of a verbosity penalty, shedding light on the underlying challenges of model behavior and performance.In this clip
From this podcast
Interconnects Audio
OpenAI's Model (behavior) Spec, RLHF transparency, and personalization questions
Related Questions