Category intelligence

Research Briefing — February 4, 2026

1225 current items analyzed and ranked.

Executive synthesis

Research Summary

Today's research features major theoretical breakthroughs alongside practical infrastructure and safety advances. The hallucination rate-distortion theorem proves factual errors are information-theoretically optimal under memory constraints—a fundamental reframing of the problem.

Kimi K2.5 releases as open-source multimodal agentic model with Agent Swarm framework achieving state-of-the-art results
Simple role conditioning reduces unsafe outputs on WildJailbreak from 81.4% to 3.6% without any training
Constant-cost self-attention via symmetric Taylor approximation could transform long-context efficiency if validated
Identity Bridge challenges the reversal curse as fundamental limitation of autoregressive models

Theoretical contributions span tropical geometry analysis proving Top-k MoE routing equivalent to combinatorial depth, first PPO convergence proof, and Ω(n) lower bounds on chain-of-thought token complexity. BLOCK-EM introduces mechanistic prevention of emergent misalignment, while SWE-Universe scales coding agent environments to 807K verified tasks.

Key Themes

AI Safety & Alignment · 83World Models & Robotics · 8LLM Reasoning & Chain-of-Thought · 14Agentic Systems & Multi-Agent · 18AI Safety and Alignment · 16LLM Efficiency & Infrastructure · 10LLM Training and Fine-tuning · 14Language Model Reasoning · 12LLM Reasoning & Self-Improvement · 12Reasoning & Chain-of-Thought · 29

Primary evidence

Top Ranked Signals

Research arXiv (Machine Learning) Feb 4

Kimi K2.5: Visual Agentic Intelligence

By Kimi Team, Tongtong Bai, Yifan Bai, Yiping Bao, S. H. Cai, Yuan Cao, Y. Charles, H. S. Che, Cheng Chen, Guanduo Chen, Huarong Chen, Jia Chen, Jiahao Chen, Jianlong Chen, Jun Chen, Kefan Chen, Liang Chen, Ruijue Chen, Xinhao Chen, Yanru Chen, Yanxu Chen, Yicun Chen, Yimin Chen, Yingjiang Chen, Yuankun Chen, Yujie Chen, Yutian Chen, Zhirong Chen, Ziwei Chen, Dazhi Cheng, Minghan Chu, Jialei Cui, Jiaqi Deng, Muxi Diao, Hao Ding, Mengfan Dong, Mengnan Dong, Yuxin Dong, Yuhao Dong, Angang Du, Chenzhuang Du, Dikang Du, Lingxiao Du, Yulun Du, Yu Fan, Shengjun Fang, Qiulin Feng, Yichen Feng, Garimugai Fu, Kelin Fu, Hongcheng Gao, Tong Gao, Yuyao Ge, Shangyi Geng, Chengyang Gong, Xiaochen Gong, Zhuoma Gongque, Qizheng Gu, Xinran Gu, Yicheng Gu, Longyu Guan, Yuanying Guo, Xiaoru Hao, Weiran He, Wenyang He, Yunjia He, Chao Hong, Hao Hu, Jiaxi Hu, Yangyang Hu, Zhenxing Hu, Ke Huang, Ruiyuan Huang, Weixiao Huang, Zhiqi Huang, Tao Jiang, Zhejun Jiang, Xinyi Jin, Yu Jing, Guokun Lai, Aidi Li, C. Li, Cheng Li, Fang Li, Guanghe Li, Guanyu Li, Haitao Li, Haoyang Li, Jia Li, Jingwei Li, Junxiong Li, Lincan Li, Mo Li, Weihong Li, Wentao Li, Xinhang Li, Xinhao Li, Yang Li, Yanhao Li, Yiwei Li, Yuxiao Li, Zhaowei Li, Zheming Li, Weilong Liao, Jiawei Lin, Xiaohan Lin, Zhishan Lin, Zichao Lin, Cheng Liu, Chenyu Liu, Hongzhang Liu, Liang Liu, Shaowei Liu, Shudong Liu, Shuran Liu, Tianwei Liu, Tianyu Liu, Weizhou Liu, Xiangyan Liu, Yangyang Liu, Yanming Liu, Yibo Liu, Yuanxin Liu, Yue Liu, Zhengying Liu, Zhongnuo Liu, Enzhe Lu, Haoyu Lu, Zhiyuan Lu, Junyu Luo, Tongxu Luo, Yashuo Luo, Long Ma, Yingwei Ma, Shaoguang Mao, Yuan Mei, Xin Men, Fanqing Meng, Zhiyong Meng, Yibo Miao, Minqing Ni, Kun Ouyang, Siyuan Pan, Bo Pang, Yuchao Qian, Ruoyu Qin, Zeyu Qin, Jiezhong Qiu, Bowen Qu, Zeyu Shang, Youbo Shao, Tianxiao Shen, Zhennan Shen, Juanfeng Shi, Lidong Shi, Shengyuan Shi, Feifan Song, Pengwei Song, Tianhui Song, Xiaoxi Song, Hongjin Su, Jianlin Su, Zhaochen Su, Lin Sui, Jinsong Sun, Junyao Sun, Tongyu Sun, Flood Sung, Yunpeng Tai, Chuning Tang, Heyi Tang, Xiaojuan Tang, Zhengyang Tang, Jiawen Tao, Shiyuan Teng, Chaoran Tian, Pengfei Tian, Ao Wang, Bowen Wang, Chensi Wang, Chuang Wang, Congcong Wang, Dingkun Wang, Dinglu Wang, Dongliang Wang, Feng Wang, Hailong Wang, Haiming Wang, Hengzhi Wang, Huaqing Wang, Hui Wang, Jiahao Wang, Jinhong Wang, Jiuzheng Wang, Kaixin Wang, Linian Wang, Qibin Wang, Shengjie Wang, Shuyi Wang, Si Wang, Wei Wang, Xiaochen Wang, Xinyuan Wang, Yao Wang, Yejie Wang, Yipu Wang, Yiqin Wang, Yucheng Wang, Yuzhi Wang, Zhaoji Wang, Zhaowei Wang, Zhengtao Wang, Zhexu Wang, Zihan Wang, Zizhe Wang, Chu Wei, Ming Wei, Chuan Wen, Zichen Wen, Chengjie Wu, Haoning Wu, Junyan Wu, Rucong Wu, Wenhao Wu, Yuefeng Wu, Yuhao Wu, Yuxin Wu, Zijian Wu, Chenjun Xiao, Jin Xie, Xiaotong Xie, Yuchong Xie, Yifei Xin, Bowei Xing, Boyu Xu, Jianfan Xu, Jing Xu, Jinjing Xu, L. H. Xu, Lin Xu, Suting Xu, Weixin Xu, Xinbo Xu, Xinran Xu, Yangchuan Xu, Yichang Xu, Yuemeng Xu, Zelai Xu, Ziyao Xu, Junjie Yan, Yuzi Yan, Guangyao Yang, Hao Yang, Junwei Yang, Kai Yang, Ningyuan Yang, Ruihan Yang, Xiaofei Yang, Xinlong Yang, Ying Yang, Yi Yang, Yi Yang, Zhen Yang, Zhilin Yang, Zonghan Yang, Haotian Yao, Dan Ye, Wenjie Ye, Zhuorui Ye, Bohong Yin, Chengzhen Yu, Longhui Yu, Tao Yu, Tianxiang Yu, Enming Yuan, Mengjie Yuan, Xiaokun Yuan, Yang Yue, Weihao Zeng, Dunyuan Zha, Haobing Zhan, Dehao Zhang, Hao Zhang, Jin Zhang, Puqi Zhang, Qiao Zhang, Rui Zhang, Xiaobin Zhang, Y. Zhang, Yadong Zhang, Yangkun Zhang, Yichi Zhang, Yizhi Zhang, Yongting Zhang, Yu Zhang, Yushun Zhang, Yutao Zhang, Yutong Zhang, Zheng Zhang, Chenguang Zhao, Feifan Zhao, Jinxiang Zhao, Shuai Zhao, Xiangyu Zhao, Yikai Zhao, Zijia Zhao, Huabin Zheng, Ruihan Zheng, Shaojie Zheng, Tengyang Zheng, Junfeng Zhong, Longguang Zhong, Weiming Zhong, M. Zhou, Runjie Zhou, Xinyu Zhou, Zaida Zhou, Jinguo Zhu, Liya Zhu, Xinhao Zhu, Yuxuan Zhu, Zhen Zhu, Jingze Zhuang, Weiyu Zhuang, Ying Zou, Xinxing Zu

88 score

AI Analysis

Previously covered in Research, Kimi K2.5 is an open-source multimodal agentic model with joint text-vision training and Agent Swarm framework for parallel task decomposition, achieving SOTA across coding, vision, reasoning, and agentic tasks.

arXiv:2602.02276v1 Announce Type: cross Abstract: We introduce Kimi K2.5, an open-source multimodal agentic model designed to advance general agentic intelligence. K2.5 emphasizes the joint optimization of text and vision so that two modalities enhance each other. This includes a series of techniques such as joint text-vision pre-training, zero-vision SFT, and joint text-vision reinforcement learning. Building on this multimodal foundation, K2.5 introduces Agent Swarm, a self-directed parallel

Multimodal ModelsAI AgentsOpen SourceVision-Language Models

Research arXiv (Artificial Intelligence) Feb 4

Simple Role Assignment is Extraordinarily Effective for Safety Alignment

By Zhou Ziheng, Jiakun Ding, Zhaowei Zhang, Ruosen Gao, Yingnian Wu, Demetri Terzopoulos, Yipeng Kang, Fangwei Zhong, Junqi Wang

85 score

AI Analysis

Proposes role conditioning as compact alternative to principle-based alignment, reducing unsafe outputs on WildJailbreak from 81.4% to 3.6% with DeepSeek-V3 through training-free role-conditioned generation and iterative role-based critics.

arXiv:2602.00061v1 Announce Type: cross Abstract: Principle-based alignment often lacks context sensitivity and completeness. Grounded in Theory of Mind, we propose role conditioning as a compact alternative: social roles (e.g., mother, judge) implicitly encode both values and the cognitive schemas required to apply them. We introduce a training-free pipeline featuring a role-conditioned generator and iterative role-based critics for refinement. Across five model families, our approach consiste

AI SafetyLLM AlignmentRole-Playing

Research arXiv (Artificial Intelligence) Feb 4

Training LLMs with Fault Tolerant HSDP on 100,000 GPUs

By Omkar Salpekar, Rohan Varma, Kenny Yu, Vladimir Ivanov, Yang Wang, Ahmed Sharif, Min Si, Shawn Xu, Feng Tian, Shengbao Zheng, Tristan Rice, Ankush Garg, Shangfu Peng, Shreyas Siravara, Wenyin Fu, Rodrigo de Castro, Adithya Gangidi, Andrey Obraztsov, Sharan Narang, Sergey Edunov, Maxim Naumov, Chunqiang Tang, Mathew Oldham

85 score

AI Analysis

Previously covered in Research, Proposes Fault Tolerant HSDP for training LLMs on 100K GPUs, using data parallel replicas as fault tolerance units with novel FTAR protocol enabling continued training during failures.

arXiv:2602.00277v1 Announce Type: cross Abstract: Large-scale training systems typically use synchronous training, requiring all GPUs to be healthy simultaneously. In our experience training on O(100K) GPUs, synchronous training results in a low efficiency due to frequent failures and long recovery time. To address this problem, we propose a novel training paradigm, Fault Tolerant Hybrid-Shared Data Parallelism (FT-HSDP). FT-HSDP uses data parallel replicas as units of fault tolerance. When f

Large-scale TrainingDistributed SystemsLLM InfrastructureFault Tolerance

Research arXiv (Artificial Intelligence) Feb 4

Hallucination is a Consequence of Space-Optimality: A Rate-Distortion Theorem for Membership Testing

By Anxin Guo, Jingwei Li

85 score

AI Analysis

Proves hallucination is information-theoretically optimal behavior under memory constraints via rate-distortion theorem for membership testing. Shows optimal models must hallucinate on non-facts even with perfect training.

arXiv:2602.00906v2 Announce Type: cross Abstract: Large language models often hallucinate with high confidence on "random facts" that lack inferable patterns. We formalize the memorization of such facts as a membership testing problem, unifying the discrete error metrics of Bloom filters with the continuous log-loss of LLMs. By analyzing this problem in the regime where facts are sparse in the universe of plausible claims, we establish a rate-distortion theorem: the optimal memory efficiency is

HallucinationInformation TheoryTheoretical MLLLM Understanding

Research arXiv (Machine Learning) Feb 4

An Approximate Ascent Approach To Prove Convergence of PPO

By Leif Doering, Daniel Schmidt, Moritz Melcher, Sebastian Kassing, Benedikt Wille, Tilman Aach, Simon Weissmann

83 score

AI Analysis

Provides first convergence proof for PPO by interpreting its policy update scheme as approximated policy gradient ascent, controlling bias from surrogate gradients using random reshuffling techniques.

arXiv:2602.03386v1 Announce Type: new Abstract: Proximal Policy Optimization (PPO) is among the most widely used deep reinforcement learning algorithms, yet its theoretical foundations remain incomplete. Most importantly, convergence and understanding of fundamental PPO advantages remain widely open. Under standard theory assumptions we show how PPO's policy update scheme (performing multiple epochs of minibatch updates on multi-use rollouts with a surrogate gradient) can be interpreted as appr

Reinforcement LearningPPOTheoretical RL

Research arXiv (Artificial Intelligence) Feb 4

Breaking the Reversal Curse in Autoregressive Language Models via Identity Bridge

By Xutao Ma, Yixiao Huang, Hanlin Zhu, Somayeh Sojoudi

82 score

AI Analysis

Previously covered in Research, Challenges the prevailing view that the reversal curse (inability to deduce B→A from training on A→B) is a fundamental limit of autoregressive LLMs. Proposes an Identity Bridge method to mitigate this limitation through slight modifications.

arXiv:2602.02470v1 Announce Type: new Abstract: Autoregressive large language models (LLMs) have achieved remarkable success in many complex tasks, yet they can still fail in very simple logical reasoning such as the "reversal curse" -- when trained on forward knowledge data of the form "$A \rightarrow B$" (e.g., Alice's husband is Bob), the model is unable to deduce the reversal knowledge "$B \leftarrow A$" (e.g., Bob's wife is Alice) during test. Extensive prior research suggests that this fa

Language ModelsReasoningModel Architecture

Research arXiv (Artificial Intelligence) Feb 4

Self-Attention at Constant Cost per Token via Symmetry-Aware Taylor Approximation

By Franz A. Heinsen, Leo Kozachkov

82 score

AI Analysis

Previously covered in Research, Derives self-attention formulation with constant cost per token by decomposing Taylor expansion into symmetric tensor product chains, achieving orders-of-magnitude reductions in memory and compute.

arXiv:2602.00294v1 Announce Type: cross Abstract: The most widely used artificial intelligence (AI) models today are Transformers employing self-attention. In its standard form, self-attention incurs costs that increase with context length, driving demand for storage, compute, and energy that is now outstripping society's ability to provide them. To help address this issue, we show that self-attention is efficiently computable to arbitrary precision with constant cost per token, achieving order

Efficient TransformersAttention MechanismsLLM Efficiency

Research arXiv (Artificial Intelligence) Feb 4

BLOCK-EM: Preventing Emergent Misalignment by Blocking Causal Features

By Muhammed Ustaomeroglu, Guannan Qu

82 score

AI Analysis

Previously covered in Research, Proposes BLOCK-EM to prevent emergent misalignment by identifying and constraining internal features that control misaligned behavior during fine-tuning. Achieves 95% reduction in misalignment across six domains.

arXiv:2602.00767v1 Announce Type: cross Abstract: Emergent misalignment can arise when a language model is fine-tuned on a narrowly scoped supervised objective: the model learns the target behavior, yet also develops undesirable out-of-domain behaviors. We investigate a mechanistic approach to preventing emergent misalignment by identifying a small set of internal features that reliably control the misaligned behavior and then discouraging the model from strengthening these features during fine

AI SafetyEmergent MisalignmentMechanistic InterpretabilityAlignment

Research arXiv (Artificial Intelligence) Feb 4

SWE-Universe: Scale Real-World Verifiable Environments to Millions

By Mouxiang Chen, Lei Zhang, Yunlong Feng, Xuwu Wang, Wenting Zhao, Ruisheng Cao, Jiaxi Yang, Jiawei Chen, Mingze Li, Zeyao Ma, Hao Ge, Zongmeng Zhang, Zeyu Cui, Dayiheng Liu, Jingren Zhou, Jianling Sun, Junyang Lin, Binyuan Hui

82 score

AI Analysis

Introduces SWE-Universe, a framework for automatically constructing 807K+ real-world software engineering environments from GitHub PRs using a building agent with self-verification.

arXiv:2602.02361v1 Announce Type: cross Abstract: We propose SWE-Universe, a scalable and efficient framework for automatically constructing real-world software engineering (SWE) verifiable environments from GitHub pull requests (PRs). To overcome the prevalent challenges of automatic building, such as low production yield, weak verifiers, and prohibitive cost, our framework utilizes a building agent powered by an efficient custom-trained model. This agent employs iterative self-verification an

Software Engineering AgentsBenchmark ConstructionCode GenerationLarge-Scale Datasets

Research arXiv (Machine Learning) Feb 4

Sparsity is Combinatorial Depth: Quantifying MoE Expressivity via Tropical Geometry

By Ye Su, Huayi Tang, Zixuan Gong, Yong Liu

82 score

AI Analysis

First theoretical analysis of Mixture-of-Experts through tropical geometry, proving that Top-k routing is algebraically isomorphic to k-th elementary symmetric tropical polynomial. Shows 'sparsity is combinatorial depth' with capacity scaling by binomial coefficient.

arXiv:2602.03204v1 Announce Type: new Abstract: While Mixture-of-Experts (MoE) architectures define the state-of-the-art, their theoretical success is often attributed to heuristic efficiency rather than geometric expressivity. In this work, we present the first analysis of MoE through the lens of tropical geometry, establishing that the Top-$k$ routing mechanism is algebraically isomorphic to the $k$-th elementary symmetric tropical polynomial. This isomorphism partitions the input space into

Mixture of ExpertsTheoretical MLArchitecture Analysis

Research arXiv (Machine Learning) Feb 4

Reasoning about Reasoning: BAPO Bounds on Chain-of-Thought Token Complexity in LLMs

By Kiran Tomlinson, Tobias Schnabel, Adith Swaminathan, Jennifer Neville

82 score

AI Analysis

Proves Ω(n) lower bounds on chain-of-thought tokens for binary majority, triplet matching, and graph reachability in BAPO model, with matching upper bounds.

arXiv:2602.02909v1 Announce Type: cross Abstract: Inference-time scaling via chain-of-thought (CoT) reasoning is a major driver of state-of-the-art LLM performance, but it comes with substantial latency and compute costs. We address a fundamental theoretical question: how many reasoning tokens are required to solve a problem as input size grows? By extending the bounded attention prefix oracle (BAPO) model--an abstraction of LLMs that quantifies the information flow required to solve a task--we

ReasoningTheoryChain-of-ThoughtComputational Complexity

Research arXiv (Computation and Language) Feb 4

Accelerating Scientific Research with Gemini: Case Studies and Common Techniques

By David P. Woodruff, Vincent Cohen-Addad, Lalit Jain, Jieming Mao, Song Zuo, MohammadHossein Bateni, Simina Branzei, Michael P. Brenner, Lin Chen, Ying Feng, Lance Fortnow, Gang Fu, Ziyi Guan, Zahra Hadizadeh, Mohammad T. Hajiaghayi, Mahdi JafariRaviz, Adel Javanmard, Karthik C. S., Ken-ichi Kawarabayashi, Ravi Kumar, Silvio Lattanzi, Euiwoong Lee, Yi Li, Ioannis Panageas, Dimitris Paparas, Benjamin Przybocki, Bernardo Subercaseaux, Ola Svensson, Shayan Taherijam, Xuan Wu, Eylon Yogev, Morteza Zadimoghaddam, Samson Zhou, Vahab Mirrokni

82 score

AI Analysis

Google researchers present case studies demonstrating successful collaboration with Gemini models to solve open problems, refute conjectures, and generate proofs across theoretical CS, economics, and physics.

arXiv:2602.03837v1 Announce Type: new Abstract: Recent advances in large language models (LLMs) have opened new avenues for accelerating scientific research. While models are increasingly capable of assisting with routine tasks, their ability to contribute to novel, expert-level mathematical discovery is less understood. We present a collection of case studies demonstrating how researchers have successfully collaborated with advanced AI models, specifically Google's Gemini-based models (in part

AI for ScienceMathematical ReasoningGoogle/DeepMind