Microsoft’s Eric Horvitz and guests discuss the Paraphrase Project, a two-year red-teaming effort that uncovered and patched a significant biosecurity vulnerability, demonstrating a model for responsibly managing the dual-use risks of generative AI in protein design.
A summary of the discussion with Dan Wang, author of "Breakneck", comparing the US and China through the lens of engineering versus legal mindsets. The conversation explores differences in infrastructure, industrial policy, manufacturing scale, and foreign policy, arguing for a nuanced view of a long-term competition rather than a short-term race.
Albert Cheng, who has led growth at Duolingo, Grammarly, and Chess.com, shares his framework for finding and scaling growth opportunities. He discusses the explore-exploit model, the keys to consumer subscription success like retention and resurrected users, and how AI is accelerating the experimentation cycle.
A reflection on Richard Sutton's "Bitter Lesson," arguing that while his critique of LLMs' inefficiency and lack of continual learning is valid, imitation learning is a complementary and necessary precursor to true reinforcement learning, much like fossil fuels were to renewable energy.
Dr. Ilia Shumailov, former DeepMind AI Security Researcher, explains why traditional security fails for AI agents. He details the unique threat model of agents, the dangers of supply chain attacks and architectural backdoors, and proposes a system-level solution called CAML to enforce security policies by design, separating model reasoning from data execution.
Dr. Ewelina Kurtys of FinalSpark explains their pioneering work in building biocomputers from living human neurons, which are up to one million times more energy-efficient than traditional silicon chips. The conversation covers the technology of reprogramming skin cells into neurons, the company's growth strategy, and the profound ethical and philosophical questions, such as potential 'Matrix' scenarios, that arise from merging biology with AI.