My favourite talks at YC AI Startup School:
1. Karpathy on LLMs as software 3.0 and a type of orchestrator OS with tools
2. Chollet on building AI that can reason - use DL to learn approx repres (intuition) to constrain discrete program search
LLMs (GPT-X) are not AGI. They have learnt pretty good representations of the training data (internet), but they have limited multi-modality, and are not able to flexibly identify and learn skills to achieve sub-goals to attain main goals, in sample-efficient ways.