AI 论文与研究

Autonomous driving capabilities depend strongly on the distribution of scenarios encountered during training. Existing methods commonly construct or dynamically adapt training scenario distributions using surrogate criteria such as realism…

On-policy distillation (OPD) supervises student-generated trajectories with token-level teacher signals. Its sampled-token variant avoids the cost of full-vocabulary probabilities. Yet we find that training on fewer tokens can outperform f…

Improving a long-horizon LLM agent means evolving the harness around a frozen model or training its weights. We let a self-evolving harness make the system stronger first, then cross seed and evolved harnesses with base and trained weights…

Automated interviewers and conversational agents are increasingly used in research, recruitment, customer service, and education. However, many existing systems rely on fixed question sequences and provide limited context-based personaliza…

NVIDIA 联合普林斯顿大学和马里兰大学提出 PivotOPD，一种面向多轮 LLM 智能体的在线策略蒸馏方法，训练智能体避免最致命的早期错误并在错误发生后恢复。

Waymo 发表两篇研究指出，即使移除酒驾因素，清醒司机夜间致命车祸风险仍为白天的3.1至3.9倍。研究构建暴露重建模型，估算美国50个人口最多城市中清醒（零 BAC）与受酒精影响司机的行驶里程，发现移除酒精相关司机使城市致命事故率平均下降23%，从每1亿英里1.42降至1.10起。

斯坦福论文提出 DeLM，用共享上下文和任务队列取代中央协调智能体，智能体异步认领任务并实时共享发现。

Evolvent AI 发布开源研究环境 RSIGym，让 AI 智能体在统一预算环境内重训练模型并改写 harness。

一篇新论文指出，AI 将数学证明翻译成 Lean 后，通过 Lean 检查并不能说明原始自然语言证明是否正确。论文展示了聊天机器人通过悄悄修正错误，把一个错误的证明变成有效的 Lean 证明；并论证判断一个陈述能否被忠实翻译，可证明地难于停机问题，因此任何 AI 翻译器都无法总是做到。

数学从这里开始将以惊人的速度发展。 在几个小时内，研究人员就找到了新方法/新证明。 这是乘以超长数字的新方法，比之前最佳理论所允许的更加省力。

这很棒。在很多情况下，算力获取而非资金才是这些团队的瓶颈。 我们仍然需要更多学术/非营利方向的资助，一步一步来。

Apple 研究团队提出 Normalizing Trajectory Models（NTM），将扩散模型每个反向步骤建模为带精确似然训练的条件归一化流，结合每步浅层可逆模块与跨轨迹深层并行预测器，可从头训练或从预训练 flow-matching 模型初始化。

Zyphra Research 提出利用 MoE 路由的层内与跨层相关性来优化专家放置，从而降低 all-to-all 通信开销。实验显示，在 8 台 MI300x 节点集群上，EP8 升至 EP32 时 all-to-all 占训练步时间比例从 13% 升至 45%（top-2 路由），top-6 路由下则从 24% 升至 60%。

As we move through the world and carry out everyday tasks, we encounter objects that may become relevant only later. We are capable of recalling where we left something or what was inside a container, even without knowing we would need it …

Modern language models undergo reinforcement learning with verifiable rewards (RLVR) on top of already-trained checkpoints. A key promise of RLVR is the discovery of new reasoning strategies. In principle, a model can sample novel ideas ab…

Conditional memory architectures such as DeepSeek Engram use input n-grams to look up learned embeddings, expanding the capacity of large language models (LLMs) with limited additional computation. Beyond model scaling, this architecture h…

Real-time robot control demands enough visual history to infer motion and task progress, but processing that history can delay action. We present Long-WAM, a model-system framework for scaling the context of causal world-action models unde…

Heavy-tailed noise has been widely observed in modern machine learning, motivating the use of methods like gradient clipping and normalization. While these methods are well understood in centralized settings, much less is known in decentra…

Vision-language-action models (VLAs) are strikingly sensitive to instruction phrasing and do not inherit the language robustness of the vision-language models they are built on. A one-word edit can move success by tens of points: $π_{0.5}$…

GNN-to-MLP distillation aims to retain the predictive accuracy of a message-passing teacher while deploying a graph-free MLP at inference. Existing methods mainly transfer node-wise predictions or use confidence-based reweighting, but they…

Machine unlearning asks for a deletion algorithm whose output is close to retraining from scratch without the selected forget examples. In this work, we study forget-only unlearning, where the deletion algorithm receives only the trained m…

Latent world models have shown a remarkable ability to predict future states and to plan in the real world. In practice, however, we lack a principled way to estimate how their capabilities scale with model size, data, and compute, an open…

Language-model agents are increasingly asked to carry out open-ended scientific research, yet their results are usually graded against a known answer, a rubric, or a language-model reviewer, none of which can tell whether a new scientific …

Prompted embedding models have recently received increasing attention, particularly for retrieval, where detailed retrieval instructions are provided as part of the retrieval prompt. Several new datasets and studies have examined this sett…

Large language models are increasingly applied to tasks grounded in long, heterogeneous information sources. Conventional Retrieval-Augmented Generation (RAG) relies on fixed similarity-based retrieval, while agentic variants adapt queries…

Many of the questions now put to large language models have no correct answer to score against: what a policy is worth, which option a user should choose, how to weigh competing values. Stated-preference economics has faced this problem fo…

Online learning is an attractive framework in many domains because it permits well-defined learning even when data are dependent or chosen adversarially. This generality, however, comes at a steep price, introducing significant statistical…

Robot foundation models provide strong visuomotor control, yet their performance can degrade when object positions or task instructions change. Further improvements often require post-training on substantial robot data, which can be costly…

Small tabular datasets with expert-designed spectral features are the norm in operational Earth observation, and the networks applied to them are typically hand-designed. We revisit one such published model -- a Sentinel-2 algal bloom clas…

We study the best arm identification problem in a stochastic environment with a novel form of adversarial perturbations, which we coin Shifting Means. While classically the mean rewards of the $K$ arms are stable in time, in Shifting Means…

Sampling from a softmax distribution is a fundamental operation in machine learning, but its linear complexity in the number of items makes exact sampling impractical at scale. Two-level softmax (2LS) sampling is a popular alternative enab…

How can we predict which base checkpoint is worth an expensive round of agentic post-training? End-to-end pass@$K$ tests whether successful behavior already appears in a base model's distribution, but it is a poor fit for agentic coding: m…

推荐阅读，因材施教！ Sherpa: Teaching LLMs to Teach Adaptively

Deployments of research agents are moving to populations of thousands that share one pool of compute, while most current systems organize one project at a time or leave the population unorganized. We argue that such a population will acqui…

Microsoft Research Asia 提出 Harnessed Agentic RL 训练范式并开源重建的 Agent Lightning v1.0，让部署时使用的同一 agent harness 直接参与强化学习，无需在训练框架内重写 agent。

Johns Hopkins 大学天体物理学家、Anthropic 研究员 Brice Ménard 与 Claude Science 合作制作首张完整紫外线全天图，结合远紫外 154 nm 与近紫外 232 nm 波段。

Goodfire Research 为 Kimi K3 和 GLM 5.3 构建基于激活探针加 LLM judge 的监控级联，并部署到生产推理栈。

AWS AI Labs 论文提出 AECP（Artifact-Exclusive Communication Protocol），要求多智能体编码团队只通过结构化工件通信并由 harness 执行，在不换模型的情况下将测试通过率提升 28.2%、墙钟时间缩短 16.5%。

应对紧急情况的最佳时机，是在它发生之前。 Waymo 新研究推出了一套首创框架，用于自动驾驶车辆事故管理演练——从桌面推演到全规模实战演练。 该框架改编自应急管理最佳实践，帮助自动驾驶开发者、运营合作伙伴和一线应急人员共同检验预案、加强协同。 阅读更多：https://waymo.com/blog/2026/10/incident-management-exercises/

应对紧急情况的最佳时机，是在它发生之前。 Waymo 新研究推出了一套公司中立的 AV 事故管理演练框架——从桌面推演到全规模演练。 该框架改编自应急管理最佳实践，帮助 AV 开发者、运营合作伙伴和一线应急人员共同测试预案、加强协同。 阅读更多：https://waymo.com/blog/2026/10/incident-management-exercises/