Evolving better AI systems

A combination of big data, big compute and novel architectures has given us ChatGPT, Claude and Gemini. These models are capable of succeeding in increasingly challenging tasks. But the data and compute requirements are staggeringly high. The models still hallucinate and lack the ability to reason or plan. What will it take to get AI significantly beyond its current capabilities? We discuss this through a combination of ideas in mathematics, computer science and neuroscience.