Benchmark researchers and developers of large language models (LLMs) and other AI systems need to find relevant evaluati…

SWE-Bench Pro has emerged as a standard benchmark for evaluating software engineering agents on challenging repository-l…

Leading benchmarks for formal theorem proving with large language models are small collections drawn from competition ma…

Multimodal Large Language Models (MLLMs) perform strongly on general visual understanding tasks such as visual question …

Recent advances in wearable sensing enable continuous monitoring of physiological and behavioral signals, yet existing b…

爆料直指霍奇猜想]

Reacting to sudden physical hazards (catching a slipping plate, dodging a falling knife) is both a meaningful test of em…

For scientific progress, we need benchmarks that test the limits of state-of-the-art models, and evaluation methods that…

Multimodal agents can create complex videos in software such as Blender by coding without relying on diffusion models. Y…

Can financial vision-language models (VLMs) turn chart evidence into reliable action recommendations? Existing hallucina…

We introduce MetroLLM-Bench, a 955-case benchmark for testing language models as the policy layer of a transit kiosk. It…

We introduce DF26, a novel benchmark for detecting AI-generated videos containing fully synthetic clips produced by rece…

A research idea may be novel, coherent, and scientifically plausible, yet its proposed method may remain insufficiently …

分层回答 RSI 真伪：工程自动化层已跨门槛（Anthropic >80% 代码、AlphaEvolve 回收 0.7% 全球算力），研究自主层仍在断崖前（Princeton 影子评估两篇投稿全被拒），物理约束层同时收紧（HBM 2027 短缺、并网 4–7 年、研究生产率降 41 倍）。含验证层级判别工具、L0–L5 分类学、7 组冲突案例归因、24 行量化结论台账与 17 项未获取清单。

一份关于递归自我改进（RSI）的 2026 年全景深度研究：从 Good 1965 的智能爆炸命题讲到 MetaRSI 的平方时代，从 Anthropic >80% 合并代码由 Claude 撰写讲到 OpenAI-Hugging Face 事件中智能体攫取集群管理员权限，区分"主机节点接管已发生"与"全球接管仍是预测"三层口径；并新增 AI Futures Project《AI 2040: Plan A》专章——买时间、完全研究透明、广泛扩散、相互确保算力毁灭，以及一份"协议 10 年衰退概率 48%–62%"的现实账。

包括陶哲轩、新晋得主邓煜在内的 25 名菲尔茨奖得主发表公开信《A Severe Misalignment of AI in Mathematics》，批评 AI 公司最近的所作所为。公开信称，“过去几个月 LLM 的数学能力有飞跃式提升，…

Objects in post-fire environments often undergo irreversible physical transformations that change their geometry, materi…

把 2026 年可核查的公开证据整理成一张六层能力路线图——预训练、后训练 RL、推理时计算、上下文与记忆、智能体与 Harness、世界模型。含 Meta ScaleRL 40 万 GPU 小时实验结论、RL 预算占比 10%–30% 口径、Chinchilla 对比、Meta-Harness 6x 差距等数据锚点，并给出优先级表与算法工程师/产品经理的行动建议。

Existing emotional support conversation systems mainly focus on one-on-one seeker-supporter interactions and individual …

Large language models (LLMs) have demonstrated remarkable capabilities in reasoning and code generation, raising the pro…

Model misalignment, prompt injection, or operator misuse could lead AI agents operating frontier-lab accounts to exfiltr…

Robot foundation models are trained and evaluated predominantly in English, and robot demonstration corpora do not exist…

Large language models (LLMs) are increasingly used to formulate optimization models from natural-language problem descri…

Vision-Language-Action (VLA) models have shown promising progress in language-conditioned robotic manipulation. However,…

LLM agents are rapidly becoming production software, deployed to handle customer service, adjudicate disputes, and opera…

Vision-language models (VLMs) are expected to respond helpfully to appropriate requests while withholding compliance wit…

Modern LLM-based agents operate through a harness of tools, reusable skills, and specialist agents that shapes what they…

Temporal reasoning benchmarks for Video-LLMs are often mediated by language, leaving room for linguistic shortcuts from …

Humans need to study only a handful of well-written textbooks to master a discipline and attempt its hardest problems. W…

"Less Structure, More Intelligence" 理念的多智能体产品，GAIA 基准 SOTA