Chat surpasses GPT: New artificial intelligence modeled after the human brain has been developed

Scientists at Singapore -based company Sapient have developed a new AI that can reason differently than existing large language models. This system, called the Hierarchical Reasoning Model (HRM), is inspired by the way the human brain processes information at multiple levels and different timescales. According to the research, HRM was trained with only 27 million parameters and 1,000 training examples. By comparison, GPT-5 is estimated to have 3 to 5 trillion parameters. Despite this, HRM performed remarkably well, particularly on challenging tests like ARC-AGI.

USES A BINARY MODULE

What sets HRM apart is that it doesn't use the commonly used "chain of thought" method. Instead, it uses a top-level module that does abstract, slow planning and a lower-level module that does fast, detailed calculations.

This allowed HRM to perform almost flawlessly in solving complex Sudoku puzzles and finding the shortest path through mazes.

"SPECIAL IMPROVEMENT TECHNIQUE"

However, independent researchers who retested the model's performance indicated that the real difference was not the architecture itself, but the specific optimization technique used in the training process. The research has not yet been peer-reviewed, but the results suggest that AI can achieve high performance with less data and smaller models.