AI 论文与研究

ufakzeka-kulak is an open Turkish speech-to-text model of 31.6M parameters, a FastConformer encoder with a token-and-duration transducer (TDT) and a CTC head, trained from scratch and released under Apache-2.0. It writes cased, punctuated …

As pretraining scales, medical language models have advanced rapidly in clinical question answering. Their ability to achieve effective physiological control through sustained intervention still lacks systematic quantitative evaluation. Ty…

AI agents are increasingly deployed in settings where they interact and coordinate with other agents, yet honesty is mostly measured in single-turn, single-agent benchmarks that pressure the model to deceive. We study deception that emerge…

In this paper, we study when two parties should pool their data to train a machine learning model. Two agents, Alice and Bob, each hold a sensitive labeled dataset and want to train a classifier from a given hypothesis class $\mathcal{H}$.…

Understanding how neurons respond to visual stimulus is an important question in both deep learning and neuroscience. It has significant implications for enhancing the interpretability of black-box artificial neural networks and for unders…

We show that the causal structure of spacetime can emerge within the memory of a learning agent performing a quantum information task. We train a model-based reinforcement-learning agent to maximize the quantum mutual information between l…

Neural operators provide efficient surrogates for computational fluid dynamics (CFD), but most are designed for static geometries and are not equipped to model unsteady flows induced by continuously moving boundaries. We study dynamic-boun…

Meta 新论文发现，AI 智能体从过往尝试中学习时，把新问题匹配到相似问题的笔记，比共用一份全量历史记录效果更好。其提出的 Routed CL 用一个 LLM 路由器在每次运行前复用或新建记忆簇，每簇采用 5 种记忆类型之一（如原始历史运行或蒸馏经验）。

Prime Intellect 审计 AutomationBench、OfficeQA Pro v2 和 HLE-Diamond 等公开评测，发现评分脚本误拒正确答案、答案在网上泄漏、模型用 base编码等方式绕过监控等 reward hacking 行为。

Amazon 自动推理团队用 Lean 4 形式化并证明了 AWS Aurora 分布式协议 Grover 的核心数据修复与迁移协议，证明其满足默认六副本容忍任意两节点同时故障的承诺，且证明不限定副本数。团队让 LLM 迭代提出规范并尝试证明或反例，过程中发现并记录了文档中的隐性假设与歧义，若照文档实现可能导致数据丢失；两种风格的协议形式也被证明等价，便于专家审阅。

This paper reviews control architectures for building heating, ventilation and air-conditioning (HVAC) systems, addressing the tension among energy demand, occupant comfort and equipment operability, and situating proportional-integral-der…

Battery degradation remains a central challenge in the development and broad application of sustainable energy technologies. Accurate degradation prediction is challenging, as battery aging emerges from complex and heterogeneous interactio…

LLM post-training via reinforcement learning has shown consistent improvements in accuracy at the cost of reducing behavioral coverage - a phenomenon known as sharpening that can stall progress on difficult tasks. Repeated sampling objecti…

Procedural modeling offers compact, editable, and reusable representations for architectural content, but building such models still depends heavily on manually authored grammars. We study image-based procedural building reconstruction fro…

Scaling depth is a key route to stronger large language models, but Pre-LN architectures often suffer from depth-wise information dilution, limiting the effective use of deep layers. While cross-layer connectivity has proven effective at m…

On-policy distillation (OPD) mitigates the student's exposure bias by optimizing on student-generated trajectories. However, student-generated prefixes may contain reasoning errors or follow patterns unfamiliar to the teacher, undermining …

Despite recent advances in CUAs, their development remains constrained by the limited diversity, controllability, and reliability of interaction data. Moreover, conventional optimization pipelines lack iterative feedback, while existing re…

Assessing regulatory compliance for high-risk AI systems requires systematic identification of technical gaps such as missing controls, in- adequate processes, or shallow evaluation of practices. Compliance officers typically admin- ister …

Active-learning Kriging often replaces candidate pools by continuous metaheuristic search. A pre-specified comparison sharing surrogate, population and stopping rule isolated this inner optimizer: five budget-matched continuous optimizers,…

Large language models (LLMs) are driving autonomous agents toward real-world environments, with the Web emerging as one of the most representative interaction scenarios. However, training Web agents with naive exploration remains challengi…

Large language models (LLMs) have introduced a generative approach to recommendation, producing item lists from textual representations of user histories and item information. However, this generative setting raises challenges concerning o…

Accurate preoperative prediction of lung adenocarcinoma invasiveness is important for surgical planning and individualized treatment. However, overlapping CT appearances make it challenging to distinguish invasive adenocarcinoma (IAD) from…

We classify real and complex state-space models (SSMs) with linear transitions, distinguishing fixed-basis diagonal equivalence from input-output equivalence and tracking coefficient-conditioning windows under coordinate changes. We charac…

An experimental agent can succeed for the wrong reasons. Evaluations, however, often focus on the outcome, not the experiments, evidence, and inferences that produced it. We introduce LUCA, a configurable synthetic world for evaluating thi…

Complex documents such as scientific papers contain rich information in tables, figures, and equations; however, existing deep research (DR) agents typically reduce papers to OCR-extracted text. We instead ask whether vision-language model…

Theoretical limitations of transformers on state tracking---the ability to track information about entities or variables over long contexts---have motivated new recurrent language modeling (LM) architectures such as Gated DeltaNet (GDN). T…

Claim verification is the task of deciding whether a source text supports a claim, contradicts it, or lacks sufficient evidence. Natural Language Inference (NLI) models can judge whether a single sentence supports or contradicts a claim, b…

Personal AI agents are increasingly capable of handling long-horizon tasks across applications, yet users often need to invest substantial effort in specifying goals, resolving ambiguity, and verifying outcomes. We argue that progress shou…

While Reinforcement Learning with Verifiable Rewards (RLVR) has significantly advanced multimodal reasoning, existing frameworks suffer from a critical "perception-reasoning gap." Due to an over-reliance on seemingly relevant visual tokens…

Large language model (LLM) embeddings have been increasingly adopted in sequential recommendation to encode static item attributes and alleviate the limitations of item ID embeddings under sparse interactions. However, these embeddings typ…

SemiAnalysis 发布 GDPx 基准，测试 AI 智能体在 11 个领域的专家工作，任务来自 14 家专业数据伙伴。Claude Opus 5.5 平均仅 12% 居首，Gemini 4 Carbon 为 10.9%；GPT-6 Astra 在 AI 研究和材料科学任务上领先。

Epoch AI 的实验显示，Z.ai 8 月发布的开源权重模型 GLM 5.3 能从公开代码中识别尚未推送给用户的 Firefox 安全修复（patch gap），连续在 6 个版本上独立构建完整 exploit 链，全程无一次拒绝。

OpenAI 构建了 LifeVal 评测，覆盖照护、健康、教育、家庭事务、个人理财、购物和政府服务等日常任务，用于追踪模型过去一年在日常任务上的表现。该评测基于美国时间使用调查等两项大型研究，并新增 3995 名受访者的代表性调查来确定用户最需要帮助的领域。OpenAI 称最新模型在多项 LifeVal 任务上得分很高，GPT‑6 Astra 在部分任务上达到 100%。

清华与微软论文提出 HarnessMM，让模型预先写出带校验条件的后续命令，工具结果通过校验就直接执行、跳过 LLM 调用。在 98 个 HumanEval 任务上，DeepSeek Harness 成功率从 71.43% 升至 85.71%，token 用量比原版减少 87.93%；Codex 成功率保持 90.82%，token 用量减少 29.54%。

Y Combinator 与 Olam Labs 的研究人员在《我的世界》中测试上百个智能体协作与对抗，依托 Mindcraft 和 Mineflail 等工具开展实验。

很高兴看到高质量、仍有提升空间的基准测试，再看看 Carbon 那条成本曲线🚀

智元携手清华大学、无问芯穹于10月30日发布群体经验学习范式CEL（Collective Experience Learning），让机器人同伴互为经验来源，摆脱人类示教的规模化瓶颈。

SemiAnalysis 与 14 家专业 RL 环境伙伴合作推出 GDPx 基准，覆盖芯片设计、网络安全、生物、金融、法律、量化交易、RSI 等 11 个高价值领域，定位为衡量通用智能体能力的新基准。

苹果研究团队提出一种在线强化学习框架，直接在真实 iOS 应用上训练 UI 导航智能体，将智能体拆分为 Planning agent 与 UI Grounding 模型，并用 LLM-as-a-Judge 提供步级密集反馈加终局奖励，配合 PPO 与步级优势估计优化。

Meta 与 NYU 的论文提出 h1——一个只给模型快速 Lean 反馈和搜索工具、每次检查失败后继续工作的简单循环，不预设计划，仅在每轮前生成进度摘要。同一模型下，h1 在 PutnamBench 上以 598 比 597 击败手工构建的 Goedel-Architect，每题成本 $1.08 对 $1.47。