TLDR Tech / AI
Google DeepMind's Online RLHF with 10x Data Efficiency
An online RLHF algorithm incrementally updated reward and language models, achieving over 10x data efficiency gains using uncertainty modeling and information-directed exploration.
Read the full story