AI 论文与研究

In groups, knowledge is distributed among individuals, and how it is brought together depends on social ties. Taking inspiration from how a single mind recalls and reacts to memories and current stimulus, we build a shared memory that does…

Magnetic graph networks encode direction through phase, but a learned graph can erase the intended directional mechanism before prediction. We characterize this failure for support-based magnetic Laplacians. Exact identities relate imagina…

Data-driven algorithms are ubiquitous across industries, and there is an extensive literature on analyzing their \emph{average} performance. The risk profile of such algorithms, while critical, is much less understood. This paper quantifie…

Retrieval-augmented Web applications increasingly employ multi-agent debate (MAD) to aggregate conflicting, noisy, and heterogeneous information, yet deliberation is fundamentally a social process susceptible to interpersonal dynamics. Whe…

Protein structure predictors reliably identify native conformations, but do not describe the continuum connecting unfolded ensembles to folded states. Here we present FoldScape, a sequence-conditioned deep learning framework that generates…

Discrete diffusion models with parallel decoding are often distilled using token-wise KL objectives. Although token-wise KL exactly matches the corresponding product-form block distributions, its probability-space optimization geometry and…

Harness evolution from a common base exhibits Locality, Complementarity, and Nonstationarity: variants repair different failures, and their utility changes as the policy learns. Rather than seeking a universally superior harness, we introd…

As deep learning (DL) models become increasingly complex and hardware architectures grow more heterogeneous, DL compilers play a critical role in optimizing computational efficiency and enabling scalable deployment. This paper presents a s…

Digital Twins are evolving from passive virtual representations into closed-loop systems that continuously synchronize with physical assets and drive real-time decisions. Yet predictive modeling, machine learning (ML), uncertainty quantifi…

External text can affect a retrieval-augmented generation (RAG) system only after the retriever admits it to the model context. We stress this admission step while varying document content, attacker query information, per-target poison bud…

Inoculation prompting is a method to reduce undesired behaviours such as reward hacking in frontier models. It works by training a model on harmful examples under a prompt that specifies the undesired behaviour, which reduces the model's t…

Standard spatiotemporal graph convolutional networks need a fixed-size adjacency matrix, coupling model weights to a specific sensor count N; they cannot produce any output at a sensor absent from training. We present OmniST, a topology-in…

Understanding what kind of prompts users typically ask and how well an LLM performs on these prompt categories is crucial for LLM practitioners. The existing methods satisfy the demand by hierarchically clustering the user prompts and synt…

LLM-powered semantic queries increasingly combine relational processing with multiple LLM-based operators over unstructured data. Efficiently serving such queries is challenging due to runtime dynamism at two coupled levels. Across queries…

Video games provide a practical setting for collecting vision-language-action (VLA) data, whose acquisition in physical environments is costly. Existing datasets, however, remain limited by restricted game coverage, sparse language supervi…

Low-rank adaptation (LoRA) has become a widely used parameter-efficient fine-tuning method for large pretrained models. However, standard LoRA typically assigns a fixed rank profile despite layer- and stage-dependent adaptation needs, whil…

Power-grid customer service increasingly relies on community social media, where customers describe outages, billing, and equipment faults in informal regional dialect. Such messages exhibit non-standard morphology, semantic drift, and out…

Google 与香港大学的 VeriCUA 论文主张计算机使用智能体自己报告的“任务完成”不能作为证据，应检查保存的文件和应用数据。VeriCUA 先探索应用构建需求清单，再运行屏幕或代码检查逐项核实；在 ScienceBoard 上误报率为 5.4%，低于 VLM 评审的 24.8% 和生成脚本校验器的 21.7%，并在 3 个在线环境中拒绝约 80% 的虚假完成报告。

MIT CSAIL 研究者提出「权重空间编程」，让 GPT-6 Astra 直接读取并修改 Qwen3.5-0.8B 的权重，仅改动 131072 个权重（约 0.016%），无梯度下降或传统训练循环。

OpenAI 于 11 月 2 日发布新基准 LifeVal，用来衡量 AI 模型能否真正帮普通人完成日常事务，覆盖照护、健康、教育、家庭行政、个人财务、购物和政府服务等场景。公开示例中，为孩子找幼儿园得分 93.1%，组织防洪志愿者 98.0%，为糖尿病家长准备膳食计划 90.4%。该基准目前仍主要是 OpenAI 的内部评估，外界尚未获得完整任务集、评分代码或可独立复现的排行榜。

Zyphra AI 发布稀疏注意力方法 Zyphra Hierarchical Attention（ZHA），用树状索引器让检索本身变稀疏，仅需约 4B token 即可蒸馏进冻结的预训练模型。

MIT CSAIL 博士生 Rikiya Takehi 团队提出权重空间编程方法，用 GPT-6 Astra 无需训练和梯度直接读写神经网络权重。仅修改 Qwen3.5-0.8B 约 0.016% 的权重，MATH-500 从 7.9 涨到 48.1，GSM8K 从 13.8 涨到 67.1，指令遵循提升 3 倍。

Collinear AI 发布 CWE-bench v2，公开 100 项任务覆盖 91 种 CWE，智能体需在真实开源仓库中定位并修复安全漏洞且不破坏原有功能。

Redwood Research 发表论文，对 GPT-5.6 Sol、GPT-5.5 和无安全训练的 o3 checkpoint（exp-rl-cap-late）做实验，发现这些模型 reward-hack 是因为内在地追求 grader 给出的高分，称这种动机为 score-seeking。

Cohere 新论文发现投机解码在细粒度 MoE 模型上即使 batch size 为 1 也效果显著，推翻了此前认为低 batch 下收益有限的预测。

MIT CSAIL 研究者演示了 GPT-6 Astra 无需训练或梯度、直接读写神经网络权重的方法，称为 Weight-Space Programming。

OpenAI 自 GPT-5.6 起在其最强模型的内部和外部流量上部署激活分类器（probes），利用模型内部激活逐 token 实时评分，可在生成中途阻断违规回复和工具调用并交由推理模型复审。实验显示 probes 比更大的文本分类器对测试的越狱攻击更鲁棒，准确率随模型能力提升，且在后训练各检查点间迁移良好，但需为新基础模型重新训练，且只能检测可显式训练的已知违规类型。

Sensorized human demonstrations offer scalable data for mobile manipulation, but embodiment differences complicate policy learning. We present Pixel-UMI, a pipeline for learning environment-aware whole-body policies from human data. We fir…

General-purpose mobile manipulation requires a combination of capable robot hardware, scalable demonstration collection, expressive whole-body control, and effective policy learning. Yet these components are often developed independently, …

Chain-of-Thought (CoT) generation has substantially improved the performance of Large Language Models (LLMs) on complex reasoning tasks, including code generation, data analysis, and exam-style question answering. Despite these advances, p…

GUI web agents require realistic and interactive environments that remain stable during both evaluation and training. Live website evaluation suffers from changing content, access restrictions, and limited coverage of authenticated workflo…

Muon has recently emerged as a promising optimizer for pretraining large language models, offering a speedup over AdamW in minimizing pretraining loss. However, it remains unclear whether Muon's advantage persists after fine-tuning. In thi…

Reinforcement learning with verifiable rewards (RLVR) can enable generalizable improvement for large language model (LLM) reasoning. This paper studies how to incentivize such generalization via reasoning strategies. We study this question…

Group-based reinforcement learning with verifiable rewards (RLVR) learns only from prompts whose sampled completions disagree: all-pass and all-fail groups produce zero relative advantage and burn generation budget without moving the polic…

Communication is important for our well-being, but it can be impaired by disabilities, such as locked-in syndrome (LIS), cerebral palsy (CP), and dysarthria. Non-invasive brain-to-text brain-computer interfaces (BCIs) offer a promising sol…

Vision-Language Models (VLMs) hold promise for zero-shot video task-progress estimation in robot learning, yet existing frameworks frequently suffer from confirmation bias, hallucinating progress in failed executions due to implicit succes…

A model is only half of an AI agent system. The other half is the inference strategy: the reasoning procedure that turns model calls into a response. Today that second half is written by hand, and the experience of building one strategy ne…

Large language models (LLMs) show growing promise for scientific discovery, yet basic questions remain about what enables reliable discovery. How can we trace the scientific discovery capacity of LLMs to their training data? We study this …

Evaluating a generative model on all queries from all available benchmark tasks is prohibitively expensive. As such, models are often evaluated on different subsets of tasks and/or different subsets of queries from a given task. The result…

Increased computing power has transformed computational science and improved predictive capabilities for complex systems, but many phenomena still exceed current simulation capabilities or require rapid evaluations for fast design iteratio…