DeepSeek: The Open-Source AI That Changed the Game

The world of open-source AI has witnessed a massive leap forward—DeepSeek R1 and V3 are rewriting the rules of reasoning in large language models. This talk will break down what is DeepSeek and how its breakthroughs in reinforcement learning, cold-start training, and distillation are creating faster, more capable AI models that can run efficiently on everyday hardware.

In this talk we’ll cover:

• How DeepSeek empowers open-source AI—running efficiently on consumer hardware while achieving state-of-the-art reasoning performance.
• The fundamentals of fine-tuning - how it’s traditionally been done, its strengths, and its limitations.
• The "aha" moment of R1-Zero - how RL-alone unlocked self-improving reasoning behaviors without human supervision.
• Distillation for smaller models - how DeepSeek R1’s reasoning capabilities were transferred to powerful, lightweight LLMs.

This session is for all skill levels and will showcase how newer techniques are challenging OpenAI’s best models while keeping AI open, accessible, and powerful. If you care about open-source AI and the future of model training, this will be a talk you won't want to miss.