Charlie and Ted run through this week's news: a humanoid robot's 100-meter dash record at Beijing's World Humanoid Games, historical context on industrial robot safety, a 100-company consortium's statement on AI cybersecurity risk, Bill Gates's recent comments on AI oversight, OpenAI's disclosure on a large-scale agent red-teaming exercise, Stability AI's $76 million raise backed by major music labels, and Nvidia's $100 billion quarterly earnings report.
Biwei Huang, a UCSD assistant professor and founder of Aether AI, joins to discuss her company's approach to robot intelligence. Aether AI builds causal world models, systems designed to learn the physical rules and causal relationships underlying how actions affect the environment. Biwei explains how this approach helps robots generalize to new environments, and the group discusses Aether AI's focus on manipulation tasks like cleaning a table or making coffee. The conversation also touches on interpretability in AI systems, the practical considerations around robot safety and controllability, and Aether AI's priorities for the next one to two years.
Key Moments:
[00:35] Robotics and AI news roundup, including industrial robot safety history
[04:35] OpenAI shares details on a large-scale agent red-teaming exercise
[08:35] Stability AI's $76M funding round
[10:35] Nvidia's quarterly earnings report
[16:35] A look at Google's Project Aura alongside XREAL and Viture
[20:35] Biwei Huang joins and gives us an introduction to Aether AI
[23:35] How causal world models differ from correlation-based approaches
[47:35] A discussion of safety, controllability and interpretability in AI systems
[52:35] Aether AI's priorities for the next one to two years
","author_name":"Charlie Fink Productions"}