Coursera Spring Sale
40% Off Coursera Plus Annual!
Grab it
Explore DeepSeek's groundbreaking AI models in this 54-minute conference talk recorded at the AI Engineer World's Fair in San Francisco. Dive deep into the technical details of DeepSeek R1 and V3 models, examining their development through pure reinforcement learning processes and innovative four-stage training pipelines. Learn about the emergence of "reflection moments" and "aha moments" in AI reasoning capabilities, and understand the distillation strategies used to optimize model performance. The session begins with a comprehensive year-in-review of the Paper Club and future plans, then transitions into detailed discussions of DeepSeek's architecture, training methodologies, and the evolution from the original DeepSeek models to the advanced R1 and R10 versions. Gain insights into how these models achieve sophisticated reasoning through reinforcement learning, the technical implementation of distillation processes, and the practical implications for AI development. The talk concludes with community engagement opportunities and actionable next steps for practitioners interested in implementing similar approaches in their own AI projects.
