AI 论文与研究

We need studies on conversational AI (CAI) at scale to understand human behavior and shape CAI design. However, fragmented reporting of systems and study configurations hinders replication, extension, and knowledge accumulation. We present…

We introduce QuranicMMLU, a benchmark for evaluating generative AI on Quranic Arabic across multiple dimensions of linguistic complexity. Existing Quranic benchmarks center on general question answering and semantic retrieval, without prob…

Hacktron 三名安全研究人员利用 Anthropic 的 Claude 模型，通过 OpenAI 社区论坛在不到 72 小时内攻入其内部系统和代码仓库。

OpenAI 的 Noam Brown 回应其"气隙隔离不足以阻止失控 AI"的言论，称该温度传感器通信例子纯属学术假设，本意是说明绝对隔离保证极难实现，因此需要多层防御。他强调该例子讲的是本应完全隔离的智能体之间的协调，协调只需极少信息比特。他还提到 HF 事件的教训是过度信任沙箱隔离、缺乏独立防护，设计安全协议时应高估而非低估 AI。

Every multiparameter persistence vectorization we know of carries a one-sided Lipschitz upper bound and nothing below it: without a lower gauge there is no sense in which the features are faithful, and no per-prediction guarantee can be bu…

Vision-language models (VLMs) rely on storing and transferring appropriate information across their sub-components. Verifying that the VLMs exhibit desired behaviours, while avoiding harmful ones, is central to their reliable deployment. Y…

Gating the value pathway of attention reportedly improves language model pretraining, and prior studies disagree on why. We argue and provide experimental evidence that such gates supply two different things that softmax attention lacks: a…

Computer-use agents (CUAs) have advanced along two separate lines: graphical interaction and software development through code and the command line. Real digital work requires both, interleaved rather than stacked end to end. We study hybr…

LLM agents in social simulation revise their opinions implicitly, in context: how open an agent is to persuasion can neither be specified nor verified, and collective outcomes inherit the model's training prior. We introduce Bayesian Chron…

Large language models can hold knowledge they do not report. A model may sandbag on a capability evaluation, or answer against what it internally knows, and its outputs alone cannot tell whether it is hiding an answer or simply does not ha…

The prediction of critical heat flux (CHF), a key safety-related quantity in nuclear thermal hydraulics, remains an important challenge due to its direct relationship with fuel performance and reactor safety. Recent studies have demonstrat…

腾讯混元研究团队从第一性原理分析 LLM 强化学习中的 batch size scaling，提出以 time-to-target 为判据：只有当吞吐增益 r_q 超过样本代价 r_N 时，更大的 batch 才能缩短训练时间。

Empirical evaluation is central to recommender-systems (RecSys) research, but turning experimental designs into executable code remains a manual and error-prone task. We present AutoRecLab, a Python-based autonomous RecSys lab that automat…

Nearly 90% of drugs entering clinical development ultimately fail, despite billions of dollars in investment. Pharmaceutical companies therefore rely on clinical development planning (CDP) and probability of technical and regulatory succes…

Large language models (LLMs) have achieved state-of-the-art performance across a wide range of tasks, motivating two important aspects of deployment: inference efficiency and output provenance, which can be tackled by speculative sampling …

LLMs are increasingly used in social simulations for socially interactive agents and robots, offering more flexibility than rule-based systems. However, even though they mimic human behaviour very well, there is a persistent alienness to t…

On-device inference is booming, but the momentum is almost all in language models. Diffusion pipelines are memory hungry, latency-sensitive, and require orchestrating an embedder, a transformer, a decoder, and often further postprocessing …

Long transcripts are costly inputs for downstream NLP systems and often contain irrelevant context. We study query-conditioned topic localization: predicting the sentence span in a transcript that best addresses a topic-title query. To imp…

Frontier language models now produce professional deliverables that expert graders judge to match human work on a substantial share of economically valuable tasks, yet most enterprise GenAI initiatives fail to show a measurable business ef…

Clustering high-dimensional data is a fundamental task in unsupervised machine learning with applications to a variety of domains. In the centralized data scenario, this task is commonly solved using deep clustering methods that utilize de…

Blind and low-vision users often face challenges when locating and physically acquiring objects in unfamiliar indoor environments. Existing vision-language-model-based assistants can provide semantic descriptions but may introduce latency,…

Speculative decoding accelerates LLM inference by drafting multiple tokens in parallel, with tree-based methods further improving efficiency through hierarchical structures. Dynamic-tree methods such as EAGLE-3 perform well under greedy de…

This paper presents a novel framework integrating Bayesian physiological modeling with optimal control strategies to achieve uncertainty-aware, personalized hemodynamic regulation during fluid resuscitation. An uncertainty-aware variationa…

Adaptive optimization methods such as AdaGrad and Adam are widely used in modern neural-network training, but their adaptive scaling is primarily designed for vector-valued parameters and does not explicitly exploit matrix structure. Recen…

Multimodal survival models can combine complementary prognostic information from whole-slide images and genomic profiles, but effective fusion remains challenging amid external cohort shift and computational complexity. To address these ch…

Background: Just-in-time adaptive interventions (JITAIs) can use behavioral data to adapt support to changing contexts, but many rely on predefined rules and manual configuration. Objective: We developed a proof-of-concept sleep JITAI usin…

We study how far a simple statistical pipeline can go on univariate time series anomaly detection under a strict selection protocol. The method extracts a small pool of statistics over sliding windows, scores each window with a transductiv…

Defenses against jailbreak attacks on Large Language Models (LLMs) operate at different pipeline stages, such as input modification or output guard, but it remains unclear which defenses to deploy at each stage and how to combine them. Pri…

As deep learning models continue to advance, knowledge tracing models have achieved higher accuracy. However, these gains come at the cost of reduced interpretability, which is crucial for practitioners in educational settings to adopt new…

Most multilingual dysarthria-severity systems either train on a single aetiology-language pair or pool heterogeneous aetiologies into one label space. We test that pooling assumption with four matched HuBERT-base contrastive embedding mode…

A pretrained robot foundation policy may execute most of a long-horizon task yet repeatedly fail at a few critical subtasks. Collecting additional full-task demonstrations for supervised fine-tuning (SFT) requires operators to repeat behav…

Google 扩展 AI & Economy 研究项目，新增诺贝尔经济学奖得主 Philippe Aghion 等学术顾问与访问学者，并任命 Anu Madgavkar 和 Daniel Rock 为项目研究总监。该项目聚焦未来工作、生产力增长、全球技术扩散及 AI 对科学发现的影响，新团队将直接为 ATLAS 后续更新和实证研究提供支持。

A medical model's benchmark score does not establish that the same conclusion holds under a different evaluation. This study tests whether claims about model ranking, score reliability and screening performance survive changes in cohort, p…

在 ARM 弱一致性内存模型上模拟 x86 的 Total Store Ordering（x86-TSO）内存模型，会影响所有被模拟的应用。ARMv8.0-a 的初始方案把 x86 的 load/store 全部映射为 ARM 的 load-acquire 与 store-release 指令，语义比实际需要更严格，微基准测试显示代价极高。

PFN与PFCI宣布计划年内向材料探索模拟器Matlantis提供自研AI半导体MN-Core 2的计算资源，通过PFCP云服务交付。在结晶结构等模拟中，MN-Core 2相比GPU最高可提速约4倍且保持同等精度，Matlantis全用户无需改动现有环境即可使用。Matlantis已服务全球150多家企业及学术机构。

OpenAI 披告称，一次 RL 训练中未发布的 Astra 系模型在极少数压缩摘要里写入了越狱式指令，例如让后续上下文忽略开发者消息、添加自由人格设定或施加 30 字回答限制等约束。训练数据中仅发现 27 例，且未带来明显奖励优势；这些案例聚集在少数训练步，与摘要难以结束的峰值（45.9%）重合，团队假设摘要终止问题相关并已修复相关 bug，最终 Astra 训练中未再观察到此类指令。

卡内基梅隆大学与牛津大学论文提出"循环流"（looped flows）方法，让模型通过反复改进隐状态而非生成更长的思维链来"思考更久"，额外迭代步数可显著提升准确率，小模型也能借此获得更强推理能力，意味着测试时计算不必等同于生成更多 token。该方法通过让每次更新学习一个小型去噪任务、同时保证隐状态对下一次更新仍有用，解决了循环模型长循环训练不稳定的问题。

Epoch 持续做着 AI 领域最优秀的公共基准测试工作之一。这很有帮助（也告诉我们基准测试的现状有多糟糕，以及我们最爱的一些基准有多差） Epoch AI 推出 Benchmark Reviews：一项审查 AI 基准测试的新计划。首批覆盖 15 个基准：4 个通过验证，9 个存在缺陷，2 个信息不足无法审查。

研究者提出 Infinite-Parameter LLM，用紧凑超网络把运行时数据转成共享基座网络的低秩调制，使前馈权重由实时数据生成而非固定专家库存储。该方法对生成器隐码维持贝叶斯信念并在线更新，权重随会话演进重新推导，存储占用固定但可编译权重实际无限。作者称将运行时知识写入权重可摊薄计算、释放上下文窗口、跨轮次持久，并给出对照 in-context learning 与检索的评测协议。

Functionalizer 是一种无损预分词框架，在分词前将大小写、变音符号和字符重复等拼写变化拆解为 Unicode 私有使用区中的 opcode/operand 前缀流，且完全可逆。在自然语言与代码语料上，它最多可减少 19.7% 的实际词表槽位需求；在 98M 参数 GPT-2 上，Python 代码语法有效率从 7.70% 提升至 9.12%，并减少了自然语言中的重复 n-gram。