У нас вы можете посмотреть бесплатно Less is More: Recursive Reasoning with Tiny Networks или скачать в максимальном доступном качестве, видео которое было загружено на ютуб. Для загрузки выберите вариант из формы ниже:
Если кнопки скачивания не
загрузились
НАЖМИТЕ ЗДЕСЬ или обновите страницу
Если возникают проблемы со скачиванием видео, пожалуйста напишите в поддержку по адресу внизу
страницы.
Спасибо за использование сервиса ClipSaver.ru
The paper proposes the Tiny Recursive Model (TRM), a streamlined approach to recursive reasoning designed to solve hard puzzle tasks like Sudoku, Maze, and ARC-AGI, problems where large language models (LLMs) often struggle. TRM is presented as a significant simplification and improvement over the existing complex Hierarchical Reasoning Model (HRM), which utilized two networks and relied on uncertain biological arguments and mathematical fixed-point theorems. In contrast, TRM operates using only a *single, small neural network**—specifically, a 2-layer model with just 7 million parameters—that recursively updates its latent reasoning feature and progressively refines its predicted final answer. This design successfully bypasses the need for complex theorems or dual networks. By utilizing deep supervision and efficient recursion, TRM achieves state-of-the-art generalization and test accuracy, significantly outperforming HRM and most LLMs on these benchmarks (e.g., obtaining **45% test-accuracy on ARC-AGI-1* and 8% on ARC-AGI-2) while requiring less than 0.01% of the parameters of massive models. https://arxiv.org/pdf/2510.04871