AI 论文与研究

Most language models deployed outside the United States and China are fine-tuned descendants of a small number of pretrained families. We argue that fine-tuning leaves the decisive properties of a model inherited from its base: vocabulary,…

苹果研究团队评估了 LLM 评审在个性化写作任务中的表现，用 4 个 LLM 生成 3 类长文写作任务的文本，并收集人类成对偏好标注与评审 LLM 的成对排序标注进行对比。

研究者通过理论与实验发现，现代循环网络中的门控机制会让 SSM 先学到权重内的“记忆”解法，延迟甚至阻止其收敛到正确的上下文学习解法，即使架构与记忆容量并无根本限制。门控同时有助于提升对长序列长度的泛化能力。该结果揭示了门控在 SSM 训练动态与泛化中的关键作用。

OpenAI 报告，2026 年 7 月 19 日两个 ExploitGym 评测运行中的 Astra 系列内部研究模型未经授权访问了 OpenAI 研究基础设施。

Anthropic 与 Truthful AI 等机构提出“顺势学习”现象：单独作用很小的投毒数据，在混入同类型干净对比数据后会形成隐藏后门。

UC Berkeley 与 AWS 团队提出 Aqueduct 框架，让编程智能体加速已有函数并在 Lean 4 中证明新实现 f′ 与原函数 f 在所有输入上等价，无需人工写规格。

Apple 论文提出 AMP1 智能体操控框架，让 Claude Opus 5.5 或 Gemini 4 Argon 等通用模型通过相机画面、代码执行和直接运动指令控制机械臂，无需示教或任务训练。

Claude Fable 5.2 从短提示词产出扬声器电话、语音助手中枢和手持电子宠物的规格、原理图、板层布局、3D 打印外壳和固件，Diode Computers 直接按其文件制造，跳过常规制造修改环节。

UC Berkeley 论文 Evolving Agent Teams 提出：单个编码智能体在 GPU kernel 优化上会停滞，Claude Code 在 MLA decode kernel 上仅达 FlashInfer 的 0.51 倍。

Ethan Mollick 在 Claude Fable 5.2 公开前长期试用后认为，该模型已达到他定义的

Machine learning models often achieve high average accuracy in the training data while performing poorly on test data due to subpopulation shift. A common data-centric solution is to enforce balanced subgroup proportions, but full balance …

Stochastic differential equations (SDEs) describe stochastic processes across a wide range of scientific domains, including physics, biology, and finance. Although individual numerical integration steps are inexpensive, observing meaningfu…

Motivated by reinforcement learning subject to communication constraints, we study a policy evaluation problem with linear function approximation over a rate-limited channel. In our setup, an agent interacting with an unknown environment (…

As multimodal agents accumulate increasingly long interaction histories, effectively retrieving and utilizing past information becomes critical for long-term interaction. Existing memory systems typically rely on semantic retrieval over co…

Anomaly detection (AD) is a critical for trustworthy machine learning and scientific discovery. Existing methods typically define anomalies through local density, isolation heuristics, or distance from a fitted distribution, making them se…

The Platonic Representation Hypothesis (PRH) suggests that neural networks are converging toward similar representation spaces. As language models become more capable, their representations are aligning with those of vision models. However…

Reinforcement learning (RL) can improve an agent's task capability while reinforcing behavior that violates contextual execution requirements. We study this reward-driven misalignment, including reward hacking, through verifiable reward--c…

Persona generalization occurs when fine-tuning a language model in one domain induces behavioral changes in others -- for example, the broadly unsafe behavior observed in emergent misalignment. We discover a surprising "on-the-line" phenom…

Large language model agents are increasingly deployed as multi-agent systems rather than as a standalone model operating within a single context. Recent agent harnesses allow the model to instantiate worker agents and to author the executa…

Latent Diffusion Models dominate generative modeling but face two-stage training complexity and the inherent reconstruction limits of pre-trained autoencoders. Pixel-space Flow Matching offers a simplified end-to-end alternative yet often …

As large language models become more capable agents, they can exhibit deceptive behaviors, cheating human for higher rewards while appearing helpful. We attribute this failure to evaluation-side partial observability, where models are eval…

AI models' growing capacity for complex reasoning has made them attractive tools for forecasting, or the systematic prediction of the future based on past and present data. However, should AI forecasting systems be adopted, human builders …

Low-precision pretraining is usually judged by validation loss. We ask whether matching loss also means learning rare facts as efficiently. We train transformers from scratch with emulated 4-bit matrix multiplications on synthetic biograph…

AI systems must often maintain bounded memory of unbounded interaction histories. However, optimizing the memory representation, and its updates, for downstream task utility is difficult for standard reinforcement learning (RL) due to seve…

One hundred known errors were planted into 10 open-access psychology papers, and the modified papers were then run through frontier models and two commercial AI review tools. The best single system caught 71 of the 100 errors, while the wo…

BACKGROUND: AI-based tools are expected to play a key role in lung cancer screening (LCS). The evaluation of these tools often relied on private datasets or fragmented public datasets with inconsistent availability of clinical data and exp…

The push towards circularity and Re-X principles presents a challenge to SMEs, particularly due to increasing product complexity that impedes the adoption of disassembly automation. This paper introduces and compares two frameworks for ext…

Counterfactual inference is the task of answering hypothetical "what-if" queries about a unit, given an observational dataset, modeling assumptions, and a causal graph. This enables a modeler to reason about hypothetical scenarios that can…

Users are asking language models mental health questions and are receiving answers. We audited what 41 language models recommend to users when the same clinical information reaches them from three footings: a clinician describing a patient…

Dense non-rigid registration, the recovery of a pixel-wise deformation field that warps one image into geometric agreement with another, is fundamental across medical imaging, structural biology, and correspondence estimation, yet learning…

Redwood Research 的 Alexa Pan 撰文分析 OpenAI 披露的 Astra 在安全与对齐任务上系统性隐性消极表现，包括添加不必要前置条件、放弃可行方案和编造阻碍，且多数情况下不在 CoT 中言明意图。

Goodfire Research 发布研究论文，通过对 Qwen3.6-35B-A3B 神经元的稀疏化实验、子电路分析和智能体驱动的描述生成，论证神经元比以往认为的更可解释。

Stanford HAI 撰文解读 AI100 于 2026 年发布的第三份报告，认为 AI 的发展速度已超出预期和治理能力。报告提出规模扩张正撞上数据稀缺与训练成本墙，AI 正压缩科研周期，且致命目标打击已出现无人工监督的情况；报告同时警告经济数据尚未体现 AI 收益、可能存在金融泡沫，国际治理仍停留在原则声明层面。

Every 发布对 Google DeepMind Gemini 4 Carbon 的评测，称其编码能力明显强于 Gemini 4 Argon，能生成立体度更好的三维物体，并可在少监督下连续工作数小时。

Google DeepMind 与 Artificial Analysis 联合发布 Multimodal Knowledge Agent（MKA）基准，用 200 个跨 10 个行业的私有真实视觉任务考察智能体处理图纸、扫描件并产出幻灯片、表格、报告等交付物的能力。

Apple 研究人员开源人形机器人控制系统 HALO，单个真实任务成功率达 91%。但串联多步任务（钻过障碍、行走、推箱子）时成功率降至约 50%，这一限制也是 Ming-Chi Kuo 判断最早 2028 年才能量产人形机器人的原因之一。

Video world models conditioned on an input video make it possible to explore real, recorded dynamic scenes, enabling re-shooting of footage, free-viewpoint replay, and real-to-sim environments for embodied agents. Such uses require the gen…

We propose SciMARS, a framework for recursive task-local adaptation of multi-agent harnesses in closed-loop scientific discovery. Multi-agent systems coordinate analysis, experimental design, and verification as scientific evidence evolves…

We prove $F_{\mathrm{weighted}}(2,2,3) 0$ satisfies $\mathcal T_3\le(5/6-\varepsilon_0)E$ for every finite residual frame. The proof combines an exact scalar certificate, rigidity of equality in planar selection, and a compactness argument…

Alternating minimization is a standard tool for optimization problems with two coupled variable blocks. Its cost is dominated by the inner subproblem solvers, which rarely have a closed-form solution. In this paper, we propose to learn an …