Mô hình suy luận đệ quy nhỏ (TRM) vượt trội hơn các LLM lớn trong các bài toán giải đố (ARC-AGI) với số lượng tham số cực kỳ nhỏ (7M), chỉ 0.01% so với LLM. Kỹ thuật mới đầy hứa hẹn cho các bài toán phức tạp! #AI #trituenhantao #machinelearning #hocmay #recursiveReasoning #suy luậnđệquy

https://www.reddit.com/r/singularity/comments/1pacxa5/less_is_more_recursive_reasoning_with_tiny/

This article introduces a Tiny Recursive Model (TRM) with only 7M parameters that outperforms many large LLMs on reasoning benchmarks like ARC-AGI, achieving 45% accuracy on ARC-AGI-1 and 8% on ARC-AGI-2. #AIResearch #EfficientAI #RecursiveReasoning arxiv.org/abs/2510.04871

Less is More: Recursive Reason...
Less is More: Recursive Reasoning with Tiny Networks

Hierarchical Reasoning Model (HRM) is a novel approach using two small neural networks recursing at different frequencies. This biologically inspired method beats Large Language models (LLMs) on hard puzzle tasks such as Sudoku, Maze, and ARC-AGI while trained with small models (27M parameters) on small data (around 1000 examples). HRM holds great promise for solving hard problems with small networks, but it is not yet well understood and may be suboptimal. We propose Tiny Recursive Model (TRM), a much simpler recursive reasoning approach that achieves significantly higher generalization than HRM, while using a single tiny network with only 2 layers. With only 7M parameters, TRM obtains 45% test-accuracy on ARC-AGI-1 and 8% on ARC-AGI-2, higher than most LLMs (e.g., Deepseek R1, o3-mini, Gemini 2.5 Pro) with less than 0.01% of the parameters.

arXiv.org
Less is More: Recursive Reasoning with Tiny Networks introduces a Tiny Recursive Model (TRM) with only 7M parameters that outperforms many large LLMs on reasoning benchmarks like ARC-AGI, achieving 45% accuracy on ARC-AGI-1 and 8% on ARC-AGI-2 — higher than Gemini 2.5 Pro and Deepseek R1 with less than 0.01 % of their size.
A promising step toward efficient reasoning.
#AIResearch #EfficientAI #RecursiveReasoning
https://arxiv.org/abs/2510.04871