AI 论文与研究

Generative recommendation (GR) represents items as semantic identifiers (SIDs) and autoregressively generates the next item's SID from interaction history. However, most methods assign the same SID length to all items despite highly unequa…

Optimizing discrete allocation weights across competing segments of a shared budget is a fundamental challenge in resource-constrained, sequential decision-making environments. In practice, this optimization is often hindered by delayed, n…

Diffusion models have revolutionized the generation of realistic artificial images. Efficient techniques such as Low Rank Adaptation (LoRA) can modify these models for particular use cases, making them more diverse and widely usable. At th…

Multimodal transformer models have driven unprecedented performance breakthroughs in modern document understanding. However, the inherent black-box nature of these models remains a major obstacle to their safe deployment in high-stakes sec…

Learning from experience is essential for building the next generation of agents that continually expand their capabilities. Pretrained vision-language models (VLMs) already internalize rich world knowledge, yet this knowledge is rarely le…

Large language model (LLM) agents rely on memory systems to leverage long-term interactions, yet existing systems typically retrieve memories based on their relevance to the current query and use them as contextual evidence. However, relev…

When the sources retrieved for a query disagree, the right response depends on why they disagree: a question with several valid answers, a contested topic, stale information and a false source each call for different behaviour. Cattan et a…

Although the recent development of LLMs has paved the way for multi-modal interaction, minority languages face difficulties keeping pace. This work presents the first open Speech Language Model for Basque, designed for low-latency, end-to-…

Many AI systems need to find every relevant items in a large collection and then organize what they found. The standard way to do this is to tag each item with labels from a curated vocabulary and search on those labels. Those labels are u…

We propose HiLo-RD (High-to-Low Budget Representation Distillation) to improve masked diffusion language models (MDLMs) under a limited generation budget. Unlike autoregressive models, MDLMs choose both what to generate and where to commit…

Instruction-following information retrieval (IFIR) requires retrievers to rank documents by both a query and its instance-specific instructions, yet instruction-conditioned contrastive supervision remains scarce. Existing synthetic methods…

We present PREEN (Preset Regression via an Exemplar-based Editor Network), which casts color style transfer as regression onto the 29 parameters of a deterministic image editor. Given a content image and a style reference, PREEN predicts a…

Existing noise mitigation strategies for wind farms usually consist in reducing turbines power output, without accounting for the wake interactions between them and the noise emission directivity. This leads to over-conservative policies a…

Google、UCLA 和 CMU 的论文推出 GenXR-Bench，用 112 个专家编写任务评测编码智能体开发 XR 应用的能力，测试结合模拟手部与房间输入，与真实头显上的盲测专家判断吻合 96.5%。

In this study, we propose a dual-branch evolutionary framework that simultaneously performs Neural Architecture Search (NAS) and Hyperparameter Optimization (HPO) for Reinforcement Learning (RL) in Autonomous Driving (AD) tasks. While NAS …

Ai2 发布研究 Old Ideas, Novel Problems、配套 benchmark、数据、代码和自动化流水线，测试 LLM 作为研究想法新颖性裁判的可靠性。

Redwood Research 发文论证应差异化加速 AI 的概念推理能力，即类似哲学、缺乏可验证标准、依赖非形式化先验论证的任务。该机构称基础提示技巧和小规模训练可让模型在其主概念推理基准上提升相当于 3 到 12 个月前沿进展的分数，且对通用能力外溢较小。团队已从直接做概念研究转向该方向，并称有可吸纳更多人的现成议程。

阿里千问团队开源 BabelArena 基准，测试同一条 agent 工作流换语言后，工具调用、环境状态、检索与打分是否会走样，并发布论文 https://arxiv.org/abs/2609.23490。

新榜智汇本月报告显示，豆包对抖音的引用权重从5月的10.79%升至8月的31.8%，每引用3条内容就有1条来自抖音；百度6月首次将B站和好看视频纳入信源库，二者权重之和达12%。报告指出，标题关键词覆盖度与引用次数的相关系数为ρ=+0.149（抖音+0.157），高于语义相似度的+0.117，建议视频创作者标题贴合用户提问用词，并确保与目标提问词语义相似度达标。

菲尔兹奖得主、芝加哥大学教授邓煜在“知识分子”“赛先生”论坛演讲中称，AI 公司蒸馏了全人类知识，却以防蒸馏为借口拒绝公布思维链，不符合学术规范；他提议通过开源数学模型让数学界掌握自己的模型，并遵循数据使用、思维链公布和可追溯性等规范。

Cartesia 发布系列博客第一篇，报告其层级网络（H-Net）在字节级语言建模上的扩缩律结果：与严格匹配的 BPE Transformer 相比，1-stage H-Net 在验证损失上节省 1.28x 计算，下游评测约 2x，代码任务中位数 2.04x、分布外编程语言中位数达 3.10x（如 Prolog 5.37x）。

Cartesia 发布 From Tokens to Bytes 系列第二篇，通过 compute-to-loss 与 loss-to-loss 两类 scaling law 对比 Byte Transformer、BPE（含/不含 pretokenization）、BLT-like 和 H-Net 等方法。

卡内基梅隆大学与 Browser Use 合作发布长程网页智能体基准 Odysseys2，含 100 个真实网页长任务、982 项 pass/fail 检查，GPT-6 Astra 以平均分 84.6 完整完成 47 项，第二名 Opus 5.5 完成 18 项，5 个模型完成 0 项。

Google Research 提出面向持续学习的 sleep 框架，让 LLM 在活跃交互与离线记忆巩固之间循环，把脆弱的短期上下文固化为长期参数。该框架结合 MoE 参数扩展、GKD 与 RL 模仿学习，可让模型持续习得新事实、保留最长 10M tokens 的长上下文知识并提升复杂推理。

OpenAI 宣布三项涉及数学、计算机科学和物理的重大发现，包括证明局部光滑猜想、反驳指数时间假说（ETH），以及涉及 QCD 的量子场论存在性证明。

Cohere 为回应透明度呼吁，发布 Qwen3.8 异常行为的 traces、原始 token 和最小复现脚本，数据集托管在 https://huggingface.co/datasets/CohereLabs/10-2026-qwen-investigation-traces。

Google DeepMind 论文将 4 个 Gemini 4 Argon 智能体放入沙盒 Minecraft 生存世界，只给一个目标击败 Ender Dragon，未分配角色或预写游戏工具，智能体通过聊天和共享记忆自行组织分工。

Google DeepMind 论文《The Illusion of Stochasticity in LLMs》发现 LLM 无法自主完成符合目标概率的随机采样。

While language models have become dramatically more capable, retrieval still largely relies on lexical matching or similarity in fixed embedding spaces. This creates a capability asymmetry: what the retriever fails to find, even the strong…

As autonomous research systems become more capable, benchmarks must both evaluate their performance and support meaningful research progress. We introduce Auto-Research Arena, a multi-target benchmark for evaluating autonomous research sys…

Evolution strategies (ES) update a model by sampling a population of perturbed copies, scoring each one, and moving in the reward-weighted direction of the perturbations. This update is usually presented as an estimate of a smoothed gradie…

Many everyday tasks require controlling contact force as well as motion. On a humanoid, this force must often be coordinated across both hands and the whole body, and cannot be specified by kinematic trajectories alone. Therefore, we prese…

Learning algorithms are usually analyzed under the assumption that informative gradients are continuously available. But this is not always true---learning signals in biological neural networks, for example, may be sparse, local, and only …

Large language models (LLMs) show promise in specialty medicine, yet existing evaluations largely focus on isolated clinical encounters rather than tracking how clinical responsibilities evolve across a longitudinal care pathway. Here we i…

Expanding multilingual ASR systems to new languages is challenging when adaptation must preserve previously supported languages without replaying source-language training data. We study this problem in connector-based Speech LLMs, where a …

As foundation models evolve from information generation to digital work execution, agents are increasingly capable of completing complex tasks across tools, applications, and workflows. However, existing research lacks a unified perspectiv…

Human oversight is a pillar of AI safety. It is often exercised by a human in the loop (HITL) who approves critical actions taken by an AI agent. AI agents' increasing ability to bypass guardrails, though, raises the question of how to enf…

In federated learning (FL), secure aggregation enables a server to compute the sum of local model updates of $N$ clients, without revealing any information on the individual model updates to the server, beyond their sum. Different clients …

Frontier multimodal large language models (MLLMs) can control robots without robot-specific training, but how their capabilities compare and interact with post-trained vision-language-action (VLA) and world action model (WAM) policies acro…

纽约大学与得克萨斯大学奥斯汀分校的研究分析了 14 个语言模型在 120 段对话、45 次编码会话和 980 篇议论文中的输出，发现模型系统性过度使用六种罕见于人类文本的句式，如名词堆叠、名词动用和主谓分离。