Broke AI Startup Beats OpenAI? | The Deepseek Revolution

Links:

Sentient AI Meetup: https://www.meetup.com/gen-ai/events/305158672/

Deepseek: https://www.deepseek.com/

Nvidia CES Announcement: https://blogs.nvidia.com/blog/ces-2025-jensen-huang/

Simulated Snake/Worm neural network video: https://youtu.be/dqL66xD2b2s?si=iI3XeEC2yCYjNIq5

In this episode, recorded during a road trip to Tahoe, hosts Mark and Shashank dive deep

into the latest AI breakthroughs, focusing on Deepseek's remarkable achievements in AI model development.

They discuss how this Chinese company managed to create a competitive 670 billion parameter model

with just a fraction of the resources of major tech companies. The conversation explores various technical

innovations, from mixed precision training to multi-headed latent attention mechanisms, and extends to

Nvidia's latest announcements at CES 2024, including their new RTX 50 series GPUs and advancements in

robotics. Join Mark and Shashank for an engaging discussion about algorithmic improvements, computational

efficiency, and the future of AI development, all while navigating through Sacramento's nighttime streets.

Key topics covered:

- Deepseek's AI model achievements

- Technical innovations in AI model training

- Nvidia's CES 2024 announcements

- Developments in robotics and self-driving technology

- The future of AI development