I implemented the Hierarchical Reasoning Model (HRM) in PyTorch and applied it to a simple pathfinding task.HRM is inspired by multi-timescale processing in the brain: a slower H module for abstract planning and a faster L module for low-level computation. Both modules are based on self-attention and attempt to model reasoning in latent space.The repo includes:- The implementation,- A demo that generates animated GIFs where you can see the model refine its solution step by step, - Results of a small ablation study on what drives performanceThe biggest driver (both accuracy and refinement ability) is *training with more segments* (outer-loop refinement), not the H/L two-timescale split. (This lines up with the ARC Prize team's analysis).Repo: https://github.com/krychu/hrmCurious to hear thoughts - iterative refinement isn’t new, but I wonder if the “loop-in-a-loop” forward pass, or varied frequencies, might hint at a useful direction for reasoning in latent space (?)Comments URL: https://news.ycombinator.com/item?id=45198143Points: 1# Comments: 0