AI 论文与研究

论文提出 GenFirst，一种先生成后重建的训练策略，实现无需两阶段流水线的直接端到端隐空间生成建模，避免隐空间坍缩。关键洞察包括 KL 散度中的熵项对防止坍缩必不可少，以及重建与生成学习动态不对称。

论文提出 FACE-Eval，一个 5，100 样本的评测，考察偏好线索出现在用户消息还是工具返回、以直接摘要还是原始工件形式给出时，推理模型思维链的忠实度变化。

论文提出 DIEM（Dynamic Important Example Mining），一个在强化微调（RFT）全程自适应利用数据的自动化框架。它在每个优化步骤中集成梯度对齐重要性估计器（近似每个样本对策略提升的边际贡献）和受限批内重加权方案（在保持更新梯度幅度的同时最大化聚合效用）。

论文提出 Super Library Agent 问题，让 Agent 顺序生成 N 个相关应用的同时维护一个可复用组件的共享 Super Library。针对朴素顺序流程提取召回率低、依赖迁移脆弱的问题，作者采用基于代码块摘要的候选引导提取、提取前代码库整合，以及利用提取轨迹和调用图信息的上下文感知迁移。

论文提出 SafeAtlas-VL 数据集，包含 150 万训练实例，将图像、请求和响应三个层面的安全判断放在五级有序量表上，覆盖 15 个危害类别和 55 个细分子类。

研究团队发布 CoVA-SFT 数据集，包含 51.9K 样本、超 222K 多模态推理步骤，覆盖 5 种布局家族和 17 个复杂任务，并配套 1，700 条测试样本的 CoVA-Bench 基准。

论文提出 QCell，一种基于查询的模型，用于显微镜图像中重叠细胞的去重叠实例分割。方法包含实例重组模块（在潜空间分解并重组查询表示）和对比查询对齐目标，并发布新的 Organoid 数据集基准；在 ISBI2014 上比 SOTA 高出 +2.2 AP 和 +2.7 AJI，代码见 https://github.com/SlavkoPrytula/QCell

一项研究对8款领先模型进行为期一年的长时程任务测试，包括GPT-5.6 Sol和Claude Opus 4.8，其性能相较人类大幅下降。最佳配置Qwen3.7-Max with Hermes最终仅获得人类平均收益的27.3%，凸显长时程执行可靠性严重不足。

自动化对齐研究框架将 Claude Opus 4.8 封装为智能体，用 2 天自主完成研究、提出方法并训练测试其他 AI，有效缓解对齐失败。主推文指出研究范式正从人类提出对齐方法，转向人类设计探索环境、AI 大规模搜索干预方案，未来人类价值或在于提供并优化目标信号。

伯克利和得克萨斯大学推出 FreeToken，新引擎可在普通设备上运行大型开源模型。8GB 游戏本跑 35B 模型达 39.3 tokens/秒，快于生产环境中的 Codex。FreeToken 让 GPU 缓存跟随路由请求，将专家缺失率从 llama.cpp 的 62% 降至 16%，其余按 PCIe 和内存速度比例分配。

Anthropic 研究员 Chen Yueh-Han 团队发布论文，展示自动化系统在 10 个对齐基准上全部提升模型表现且不损害整体性能。系统模拟传统研究流程，搜索文献、提出方法并训练 30 分钟，迭代优化。论文称最佳 AAR 方法平均 6 小时内超越人类专家提议，成本约每小时 $4 API 推理，远低于人类研究员的 $150。

Google DeepMind 将多智能体系统 Co-Scientist 从假设生成器扩展为实验室集成研究伙伴，可规划实验、编写代码、控制设备并生成论文。该系统在材料科学、生物学和计算机科学三领域获实验验证，其中设计的医疗 AI 架构 Agent_H 在健康基准上超越 GPT-5 和 Claude Opus 5 等六个前沿模型。

人类会随着经验而进步。AI？则不然。我们在Earthborne Rangers上运行了一个人类基线--这是我们EBR-bench基准背后的复杂、长周期桌游。人类起初表现不佳，但最优秀的玩家在5次游戏后掌握了这款游戏。而我们评估过的所有AI都从未达到这一水平。

We propose QGPINNs, a physics-informed neural network framework developed in PyTorch for the numerical solution of nonlocal differential equations on quantum graphs. The framework is designed as a general computational implementation in wh…

Tendon-driven hands are anthropomorphic, and moving the actuators off the joints is what makes a hand of this capability affordable to build. Two effects produce that saving. Routing force through a cable removes the requirement that a mot…

Synthetic data can improve statistical inference when real data are scarce, but naively treating synthetic samples as real data can introduce bias and lead to unreliable inference. We develop a general framework for synthetic-augmented inf…

Anthropic 新研究显示，Claude 可自主完成对齐研究，且被纠正的模型比执行纠正的模型能力更强。Claude 自主发现的安全训练方法，超越了 28 位资深研究员各自提出的方案。五个智能体并行工作长达 48 小时，共享结果，并通过隐藏测试和能力门控筛选过拟合与明显退化。

We study the mixing time of weighted Dikin walks for sampling from exponential distributions on polytopes and truncated positive-semidefinite (PSD) cones. Our first result gives a general total-variation mixing bound under strong self-conc…

We study kernel ridge regression under anisotropic Gaussian data, where the input covariance decays as a power law with exponent $α\geq 0$ for polynomial inner-product kernels. We derive asymptotically sharp expressions for the kernel spec…

Can a listener recover what a speaker means from the form of an utterance alone? We answer this question information-theoretically, and for a listener given by any featurizer of text, including the hidden states of contemporary large langu…

Neural-network optimization in 2025-2026 is no longer well described as a succession of new Adam variants. The design space has expanded from coordinates to matrices and layers, from fixed training horizons to policies over time, and from …

Modern agent systems assemble capabilities at runtime, and this dynamic composition has recently received a complete formal treat ment in the spatiotemporal-composability calculus, in which a capability is a component carrying a tracked in…

As a precursor to high-dimensional biomedical data modeling, reliable feature selection can reduce computational expense, improve modeling performance, and yield simpler, more interpretable models. However, most filter-based feature select…

Recent generative approaches to geometry estimation adapt pretrained image diffusion models and treat the task as image-conditioned generation. Leveraging off-the-shelf image diffusion models, they either (i) train task-specific geometry m…

Model merging combines multiple task-specific fine-tuned LLMs into a single multi-task model without additional training. However, merged models are known to suffer from representation bias: systematic drift between the merged model's hidd…

A code world model accepted by a sampling gate can be exactly right on everything the gate can see and arbitrarily wrong beyond it. We characterize what a certified model can know, and what its errors can cost, when the omission is an annu…

Recent advances in generative AI allow users to create 3D models from text or images. However, these models prioritize visual plausibility over geometric accuracy, often generating results with flaws that compromise their intended use post…

Texture image classification plays a significant role in computer vision applications, including industrial inspection, medical image analysis, remote sensing, and object recognition. Handcrafted features can capture local texture characte…

一项新研究称，其 AI 系统在两位人类工程师仅提供硬件规格后，自主完成了性能模型、RTL、UVM 验证、形式化证明、固件及内核的全部设计，并在两周内部署，规格以下无人工干预。

We investigate whether automatic speech recognition (ASR) errors in user input can lead to unsafe outputs from Embodied AI (EAI) models. We find that ASR errors can lead to harmful instructions being accepted and executed by EAI models, th…

Neural operators provide fast surrogate models for approximating operators between function spaces, but their predictions often lack uncertainty quantification. We develop a split conformal framework to guarantee that a calibrated pointwis…

When training Mixture-of-Experts (MoE) language models with expert parallelism, all-to-all token dispatch and combine collectives can consume a substantial fraction of end-to-end training time. In this work, we study communication-efficien…

Forced alignment evaluation typically requires manually annotated timestamps, limiting large-scale and multilingual analysis. We introduce two corpus-level metrics based on self-supervised (SSL) speech representations for reference-free fo…

To determine the real-world effectiveness of machine learning based malware detection, it is vital to evaluate its robustness against highly capable adversaries. However, state-of-the-art attacks do not effectively model realistic adversar…

AI coding agents, software tools that automate development tasks through reasoning and tool use, are increasingly extended through plugin marketplaces, yet the structure, maintenance, and co-evolution dynamics of these emerging repositorie…

美团LongCat在36项AI研发任务上评测7个前沿模型，覆盖756条轨迹。结果显示：强优化性能不等于真正创新，252个方案中仅3个算新颖；可靠性比峰值性能更能区分模型；经验积累可能有益也可能误导，而自动化harness优化带来的收益可迁移至其他任务和模型。当前智能体更像工程优化器而非完全自主研究者。

Google DeepMind 启动首个专有前沿 AI 模型的双盲评估试点，与新加坡 AI 安全研究所等合作，在 Gemini Flash Lite 系列模型上运行。

Google 一篇论文提出将原始执行轨迹、持久知识 wiki 与可执行技能三者分离，经验先沉淀进 wiki，后续技能更新均基于该 wiki。消融实验证实 wiki 是主要增益来源：带进化技能的小模型可超越无技能的大模型，且技能可跨模型家族迁移。论文：arxiv.org/abs/2608.27454。

Google DeepMind 新论文将 Co-Scientist 从模拟带入真实实验。在计算机科学中，它发现的推理时扩展架构在盲审下击败了 HealthBench Hard 和 Professional 上的六个前沿模型。

Google 发布论文 WikiSkill，将智能体技能进化系统拆分为原始执行轨迹、持久知识 wiki 与可执行技能三部分，经验先沉淀进 wiki，后续技能更新基于该知识库。消融实验显示 wiki 是主要增益来源：小模型配进化技能可超越规模大得多的无技能模型，且一个模型进化出的技能可跨模型家族迁移，有时甚至优于自进化技能。论文地址：arxiv.org/abs/2608.27454。