This is the second part of episode 10 of Effortless Podcast, hosts Dheeraj Pandey and Amit Prakash sit down with Alex Dimakis, a renowned AI researcher and professor, to discuss one of the biggest breakthroughs in open AI models—DeepSeek R1. They explore how DeepSeek’s innovations in reasoning, reinforcement learning, and efficiency optimizations are reshaping the AI landscape. The conversation covers the shift from large, proprietary AI models to open-source alternatives, the role of post-training fine-tuning, and how reinforcement learning (GRPO) enables reasoning capabilities in LLMs. They also dive into KV caching, mixture of experts, multi-token prediction, and what this means for NVIDIA, hardware players, and AI startups.