Kanjun, Daniel, and Josh discuss the challenges of interpretability and predictability in AI models, highlighting the need for better understanding and composability. They also touch on the importance of ablative analysis and the impact of short-term pressures on research.
