HRM

This paper I read from here has an interesting mathematical model for reasoning based on cognitive science. https://arxiv.org/abs/2506.21734 (there is also code here https://github.com/sapientinc/HRM) I think we will see dramatic performance increases on "reasoning" problems when this is worked into existing AI architectures.

Skepticism warranted since ARC-AGI hasn't been reproduced and there's at least some debate about validation data leaking into the training set: https://github.com/sapientinc/HRM/issues/18 I'm eagerly awaiting reproduction. I hope the results hold up and someone finds a way to bolt language onto it.

I am still reading the paper, but it is worth noting that this is not an LLM! It is closer to something like AlphaGo, trained only on ARC, Sudoku and mazes. I am skeptical that you could add a bunch of science facts and programming examples without degrading the performance on ARC / etc - frankly it’s completely unclear to me how you would make this architecture into a chatbot, period, but I haven’t thought about it very much. Comparing the maze/Sudoku results to LLMs rather than maze/Sudoku-specific AIs strikes me as blatantly dishonest. “1k Sudoku training examples” is also dishonest, they generate about a million of them with permutations: https://news.ycombinator.com/item?id=44701264 (see also https://github.com/sapientinc/HRM/blob/main/dataset/build_su... And they seem to have deleted the Sudoku training data! Or maybe they made it p

Source code available at https://github.com/sapientinc/HRM

Yeah, the results look incredible indeed. That's why I and many others here have decided to download, review, and test the code published by the authors.[a] If their code doesn't live up to their claims, we will all ignore their work and move on. If their code lives up to their claims, no one can argue with it. In my experience, when authors publish working code, it's usually a good sign. --- [a] https://github.com/sapientinc/HRM

sapientinc/HRM: Hierarchical Reasoning Model: https://github.com/sapientinc/HRM : > [...] Inspired by the hierarchical and multi-timescale processing in the human brain, we propose the Hierarchical Reasoning Model (HRM), a novel recurrent architecture that attains significant computational depth while maintaining both training stability and efficiency. HRM executes sequential reasoning tasks in a single forward pass without explicit supervision of the intermediate process, through two interdependent recurrent modules: a high-level module responsible for slow, abstract planning, and a low-level module handling rapid, detailed computations. With only 27 million parameters, HRM achieves exceptional performance on complex reasoning tasks using only 1000 training samples. The model operates without pre-training or CoT data, yet achieves nearly

Looking at the code, there is a lot of data augmentation going on there. For the Sudoku and ARC data sets, they augment every example by a factor of 1,000. https://github.com/sapientinc/HRM/blob/main/dataset/build_ar...

The authors' code is at https://github.com/sapientinc/HRM . In fields like AI/ML, I'll take a preprint with working code over peer-reviewed work without any code, always, even when the preprint isn't well edited. Everyone everywhere can review a preprint and its published code, instead of a tiny number of hand-chosen reviewers who are often overworked, underpaid, and on tight schedules. If the authors' claims hold up, the work will gain recognition. If the claims don't hold up, the work will eventually be ignored. Credentials are basically irrelevant. Think of it as open-source, distributed, global review. It may be messy and ad-hoc, since no one is in charge, but it works much better than traditional peer review!

Yeah, that was pretty much my reaction. I will need time on a computer too. The repo is at https://github.com/sapientinc/HRM . I love it when authors publish working code. It's usually a good sign. If the code does what the authors claim, no one can argue with it!

If I understand this correctly, it learns the rules of Sudoku by looking at 1,000 examples of (puzzle, solution) pairs. It is then able to solve previously unseen puzzles with 55% accuracy. If given millions of examples, it becomes almost perfect. This is apparently without pretraining of any sort, which is kind of amazing. In contrast, systems like AlphaZero have the rules to go or chess built-in, and only learn the strategy, not the rules. Off to their GitHub repository [1] to see this for myself. [1] https://github.com/sapientinc/HRM

Nice! They provide trained checkpoints on their GitHub. Repeating their results would be a good start. https://github.com/sapientinc/HRM

24 hours on a 4070. Seems quite doable. https://github.com/sapientinc/HRM