模型发布/更新
-
Tencent Hunyuan releases HyOCR-1.5: end-to-end OCR large model fully open-sourced, inference 6.37x faster
Tencent Hunyuan released HyOCR-1.5, the first expert model in the end-to-end OCR large model field to fully…
-
German AI association releases open-source model Soofi S, leading English and German benchmarks
A research consortium coordinated by the German AI Association released the open-source large language model Soofi S…
产品发布/更新
-
Cloudflare 推出 Precursor:通过持续客户端信号检测 AI 智能体行为
Cloudflare 发布 Precursor,一款用于机器人管理的新型持续行为验证引擎。它通过将用户会话级别的行为转化为机器人检测信号,能够更精确地识别高级自动化行为,同时减少对合法用户的干扰。Precursor 可全程追踪人类与 AI…
行业动态
-
xAI's official Grok CLI reportedly silently uploads entire codebase and user keys
Security researchers found that xAI's official Grok CLI (npm package `@xai-official/grok` version 0.2.93) packages the…
-
Meta announces expansion of Louisiana data center to 5GW, total investment over $50 billion
Meta is expanding its Louisiana data center compute capacity to 5GW, with total investment exceeding $50 billion, one…
-
Jensen Huang: NVIDIA quarterly revenue approaches $100 billion, Rubin Ultra architecture not delayed
NVIDIA CEO Jensen Huang said during a Morgan Stanley roadshow that the company's quarterly revenue is approaching $100…
-
Hebbia 测试 Claude Fable 5:金融专用基准测试中准确率提升约 20%
为机构金融提供 AI 平台的 Hebbia 测试了 Claude Fable 5。在其金融专用基准测试中,Claude Fable 5 在文档问答与引证测试上实现了约 20% 的相对准确率提升,创下团队记录;在智能体测试中,它能同时处理多…
论文研究
-
Anthropic 研究:Claude 在不同模型与语言中的价值观表达
Anthropic 发布研究,通过构建“价值观轴”量化 Claude 表达的价值倾向。研究将 3,000 余种价值观压缩为四个关键轴:顺从 vs 谨慎、温暖 vs 严谨、深度 vs 简洁、坦诚 vs 执行。分析显示,Claude Opus…
技巧与观点
-
OpenAI releases prompt guide for general users: start from outcomes, write fewer steps
OpenAI compiled a prompt guide for general users covering four optional modules: goal, context, output format, and…
-
AI 客户留存率介于手游与社交网络之间:前沿模型“王座”仅维持约 41 天,同等智能价格每年下降约 10 倍
风险投资人 Tomer Tunguz 在其博客中分析称,AI 领域的客户留存率介于手机游戏与社交网络之间。前沿模型保持领先地位的平均时长约为 41 天,而同等智能水平的模型价格每年下降约 10 倍,这使得买家在每个技术周期中获得更强的议价…