Reward Model Challenges
Misha discusses the complexities of training reward models for chatbots, highlighting how a poorly designed reward system can lead to models that avoid answering questions altogether. Sonya adds that while pre-training techniques are well-established, post-training remains an evolving field, emphasizing the need for clarity in understanding the roles of both stages in AI development.In this clip
From this podcast
Training Data
Reflection AI’s Misha Laskin on the AlphaGo Moment for LLMs | Training Data
Related Questions
What challenges are faced in training large language models (LLMs) as discussed in the episode Jennifer Prendki Interview - Agile Machine Learning - TWiML Talk #46 and the clip Monitoring NLP Models?
I have a question about the episode Navigating Machine Learning Careers: Insights from Meta to Consulting // Ilya Reznik // #286 and the clip Fine Tuning Insights on how to train large language models (LLMs) accurately.