The high cost of computing power is a major bottleneck for Artificial Intelligence, presenting a clear investment opportunity. Focus on companies that make AI more efficient, particularly those in the semiconductor and software sectors. Consider investing in firms designing the next generation of powerful AI chips to meet processing demand. Additionally, look into software companies that are developing solutions to run large models more cost-effectively. These businesses are positioned to benefit as the demand for AI processing grows.

Artificial Intelligence (AI) & Robotics

• The discussion centers on a new approach to building AI for robots, moving away from using separate neural nets for specific tasks (e.g., a "dishes neural net" or a "logistics neural net").
• The guest, Brett Adcock, mentions his project uses one neural net that is trained on a massive amount of diverse data.
• This single, large model approach has shown "positive transfer," meaning the robot can generalize and learn new physical motions better because it has a broader base of knowledge. The analogy used is that learning to play the piano might make you a slightly better soccer player.
• A key challenge identified is that while data storage is cheap, "the processing is very expensive." It is inefficient to run the entire massive neural network for every simple task. This presents a "hybrid problem" of how to access only the relevant parts of the model for a specific action.

Takeaways

• The future of robotics may be dominated by companies that successfully develop a single, generalized AI model rather than those who build task-specific software libraries. Investors should pay attention to companies focused on creating these "generalist" AI brains for robots.
• A major bottleneck in the advancement of AI and robotics is the high cost of computing power. This highlights a significant investment opportunity in companies that are solving this problem.
  • This includes companies developing more efficient and powerful AI chips (hardware).
  • This also includes companies creating innovative software solutions that make running these large, complex models more cost-effective.