#python #brain_inspired_ai #deep_learning #large_language_models #reasoning

The Hierarchical Reasoning Model (HRM) is a new type of AI that reasons more like a human brain, using a fast part for quick details and a slow part for big-picture planning. It solves hard logic tasks like Sudoku, mazes, and IQ-style puzzles very well, even though it is tiny (only 27 million parameters) and learns from very little data (just 1,000 examples). Unlike most large language models, it does not need long chains of written reasoning steps or huge amounts of training, which makes it much faster, cheaper, and more efficient. For the user, this means powerful reasoning in a small, fast system that can run on ordinary hardware and still beat much larger models on tough problems.

https://github.com/sapientinc/HRM