#02 - DeepSeek

The Chinese tech start-up DeepSeek used reinforcement learning and other creative techniques to develop an ultra-efficient chatbot that was less costly to make, uses fewer chips and requires much less electrical power than better-known alternatives, such as OpenAI’s ChatGPT or Anthropic’s Claude.  To discuss the implications, Intrepid Growth Partners gathered the Derby Mill team, including the pioneer of reinforcement learning himself, Richard Sutton, as well as Sendhil Mullainathan and Niamh Gavin, all of whom are Intrepid GP senior advisors. The team joins show host and Intrepid partner Mark Shulgan along with special guest Kevin Bryan, economist and co-founder of the AI start-up, All Day TA.  Questions our hosts will discuss: - How can DeepSeek perform as well as OpenAI's most advanced technology for many tasks—but operate at a fraction of the cost? - What potential economic effects could result from DeepSeek’s open-source code and technical specifications? - What implications does DeepSeek have for investors seeking strategies to invest in the future of AI? -We’ll also make some predictions about where things go from here. LINKS Derby Mill show website: ⁠https://insights.intrepidgp.com/podcast⁠. The ⁠DeepSeek academic paper⁠ that started it all https://arxiv.org/abs/2501.12948 Find Kevin Bryan on X ⁠https://x.com/Afinetheorem and on Bluesky https://bsky.app/profile/afinetheorem.bsky.social. His website is ⁠http://kevinbryanecon.com. See Kevin Bryan’s start-up, ⁠All Day TA⁠, which provides AI assistants for university professors https://www.alldayta.com/ Rich Sutton’s ⁠home page⁠ http://incompleteideas.net/. Follow Rich on X https://x.com/RichardSSutton Sendhil Mullainathan’s ⁠website https://sendhil.org/⁠. Follow Sendhil on X https://x.com/m_sendhil?lang=en