Apple and CMU researchers introduce SSR, a framework that replaces free-form reasoning in multimodal search agents with selection from six predefined reasoning options, cutting per-step latency over 90% while matching task success rates across seven benchmarks.
A 10‑million‑parameter GRAM model from Bengio, KAIST, Mila and NYU achieves 97% accuracy on Sudoku‑Extreme and competitive scores on ARC‑AGI tasks by replacing deterministic recursive updates with a probabilistic multi‑trajectory process, and extensive ablations show that both random guidance and depth‑supervised training are essential for its performance.
The paper introduces GRAM (Generative Recursive Reasoning Models), a probabilistic multi‑trajectory recursive reasoning framework that injects learnable randomness into each recursion step, enabling parallel sampling and achieving higher accuracy than deterministic baselines across tasks such as Sudoku‑Extreme, N‑Queens, ARC‑AGI and unconditional generation.
The GRAM model introduced by Yoshua Bengio’s team replaces deterministic recursive updates with probabilistic multi‑trajectory sampling, enabling a 10 M‑parameter network to achieve 97 % accuracy on Sudoku‑Extreme, 52 %/11 % on ARC‑AGI, and near‑perfect results on N‑Queens and graph‑coloring, while also supporting unconditional generation tasks.
This article introduces five free, open‑source Model Context Protocol (MCP) services that extend AI capabilities—covering web scraping, browser automation, document‑to‑Markdown conversion, real‑time documentation, and structured reasoning—along with concise step‑by‑step setup instructions for each.
The KAG-Thinker model, a collaborative effort by Ant Group, Zhejiang University, and Tongji University, introduces a hierarchical "breadth splitting + depth solving" framework that enhances logical stability, knowledge utilization, and retrieval robustness for complex multi‑hop reasoning tasks across general and specialized domains.
The article introduces CodePlan, a novel framework that injects code‑form planning into large language model reasoning, addressing the limitations of natural‑language‑only approaches and demonstrating significant performance gains across diverse benchmark tasks.
