Dan Shipper

Dan explores OpenAI's new language model o1, which utilizes 'chain of thought' reasoning to improve performance on complex tasks. Unlike previous models that would 'blurt out' responses, o1 is trained via reinforcement learning to verbalize its step-by-step thought process. This ability to 'think out loud' keeps the AI focused, allowing it to spend more time on challenging queries. Dan speculates that investing more 'test-time compute' could become a new paradigm for enhancing AI capabilities, akin to the role of increased training data and compute power. While o1 excels at math, science, and coding problems, Dan questions whether it can truly create novel knowledge or simply recombine existing information in unprecedented ways.