A live stream conversation between Josh Starmer (StatQuest) and Luis Serrano covering DeepSeek's technical innovations, including its use of reinforcement learning over supervised fine-tuning, mixture of experts, knowledge distillation, and why it remains fundamentally a Transformer architecture. They also discuss attention mechanisms, KAN (Kolmogorov-Arnold Networks), chain-of-thought prompting, GRPO, LangChain, and upcoming educational content including Josh's deeplearning.ai short course on attention and reinforcement learning video series.
•59m watch time
14 Impressions