它是全球范围内，第一个敢说在性能上能和行业标杆OpenAI DeepResearch掰手腕，并且选择将自己彻彻底底、从头到脚全部开源的Web Agent（网络智能体）。在多个业界公认的权威基准测试中，Tongyi DeepResearch不仅表现优异 ...

China and ASEAN nations launched their first ministerial dialogue on artificial intelligence on Thursday, establishing a new ...

这项由字节跳动种子团队（ByteDance Seed）的辛然、肖霞等研究人员联合卡内基梅隆大学和北京大学开发的突破性研究发表于2025年9月，论文题目为《Scaling up Multi-Turn Off-Policy RL and ...

近日，深度求索（DeepSeek）团队的研究成果以“DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement ...

在号称人类最后的考试榜单HLE（Humanity's Last Exam）中，通义DeepResearch更是拿下了32.9%的最高分，超越DeepSeek-V3.1（29.8%）和OpenAI DeepResearch（26.6%），霸榜全球第一！

在解决推理问题时，由此诞生的 DeepSeek-R1-Zero 倾向于生成更长的回答，在每个回答中融入验证、反思和探索替代方案的过程。尽管研究人员并未明确告诉模型何为推理，也没有任何监督微调（SFT，Supervised ...

HONG KONG, Sept. 17 (Xinhua) -- John Lee, chief executive of China's Hong Kong Special Administrative Region (HKSAR), delivered his fourth policy address during his term on Wednesday, vowing to ...

此前誓言远离中国的外资正在悄然回归，且本地信号显示这不仅仅是短期的逼空。内地媒体记录到北向资金稳步回归，政策后备力量显而易见，AI—半导体复合体在承担重要推动作用。上证综指冲击十年高点、港股创四年高位，不只是标头事件；它们是协调性政策抛售保护（pol ...

在强化学习（Reinforcement Learning, ...

The U.S. Federal Reserve is widely anticipated by market observers to cut rates as early as this September, even while inflation stays firm and growth indicators weaken. U.S. equities are still near r ...