my predictions for 2025 in AI:

- agents be able to do reliably well in long-range, multi-step tasks (some crazy new architectures/planning strategies/learning strategies)

- important empirical work in mech-interp (breakthroughs in scalability) + alignment (develop kind of value alignment frameworks)

- robotics + agentic ai will take off

- oh while we are at it, "world peace" and why not indeed :)