A comprehensive index of artificial intelligence and machine-learning research with AI-generated summaries, citation metrics, and direct links to papers and code.
Subbarao Kambhampati
Argues that LLMs lack genuine reasoning and planning capabilities, despite their impressive language generation, and that their apparent success is due to memorization and pattern matching.
Romain Cosentino, Sarath Shekkizhar
This paper links LLM reasoning ability to the density of self-attention graphs, showing higher intrinsic dimension in MLP inputs implies greater expressive capacity.
Yuqing Yang, Yan Ma, Pengfei Liu
Proposes a progressive weak-to-strong reasoning framework where a strong model refines its own training data without human or advanced model input, significantly improving reasoning on GSM8K and MATH.
Yu Wang, Shiwan Zhao, Zhihu Wang, et al.
SCoT improves LLM reasoning by first eliciting a problem-solving strategy before generating Chain-of-Thought steps, achieving significant gains on reasoning benchmarks.
Zhengren Wang, Jiayang Yu, Dongsheng Ma, et al.
RARE decouples knowledge storage from reasoning by externalizing domain knowledge to retrievable sources and internalizing reasoning patterns, enabling lightweight models to surpass GPT-4 and DeepSeek-R1 by ~20% accuracy.
Annie Wong, Thomas H. W. Back, A. Plaat, et al.
Evaluates prompting strategies in dynamic environments, finding strategic prompting can close performance gaps but reveals persistent reasoning limitations in LLMs.
Yuzhong Zhao, Yue Liu, Junpeng Liu, et al.
GMPO improves GRPO stability by replacing arithmetic mean with geometric mean of token rewards, reducing outlier sensitivity and boosting reasoning performance.
Chengshuai Zhao, Zhen Tan, Pingchuan Ma, et al.
Proposes a data distribution lens to understand when and why Chain-of-Thought reasoning succeeds or fails, revealing it as a brittle mirage beyond training distributions.
Kaiwen Wei, Rui Shan, Dongsheng Zou, et al.
MIRAGE enhances test-time scaling for medical QA by combining multi-path parallel inference with structured knowledge graph retrieval to reduce error accumulation and improve traceability.
Priyanka Kargupta, S. Li, Haocheng Wang, et al.
This paper synthesizes cognitive science into a taxonomy of 28 cognitive elements, evaluates 192K LLM traces across modalities, and develops test-time guidance improving performance by up to 66.7%.
Beichen Zhang, Yuhang Zang, Xiao-wen Dong, et al.
Proposes Vision-Language Synergy Reasoning (VLSR) and Modality-Switch Self-Correction (MSSC) to improve abstract reasoning on ARC-AGI by combining visual abstraction with linguistic reasoning.
Yaoming Li, Guangxiang Zhao, Qilong Shi, et al.
First primer synthesizing over 150 studies on post-training reasoning data, organizing the field around four key questions.