Pentagon planners are weighing options, though no final decision has been made.

A deep dive into adapting language models for diffusion-based ARC solving, what went wrong, and what I learned about the fundamental tradeoffs in generative approaches.