Најди сличен содржај

EU Scrutinizes Bing's Generative AI over Potential Election Risks The European Commission is escalating its investigation into potential risks associated with generative AI features on Microsoft's Bing search engine. The Commission seeks information beyond what Microsoft previously provided. The request stems from concerns that Bing might be violating the Digital Services Act. This act outlines risk assessments and mitigation measures for online platforms, particularly regarding "hallucinations" (false information generated by AI systems), deepfakes (manipulated media), and automated manipulation of services that could mislead voters. The Commission emphasizes the potential threats posed by generative AI to civic discourse and electoral processes, particularly in the upcoming European Parliament elections. Microsoft has until May 27th to comply with the legally binding request. Non-compliance could lead to significant fines of up to 1% of Microsoft's global turnover, or daily penalty payments calculated as a percentage of average daily income. #DSA

🇦🇿 Ozarbayjonda Best Startup Project yarim final bosqichi bo‘lib o‘tdi! “Digital Startup Awards” (DSA) butun mintaqa bo‘ylab eng iqtidorli startap asoschilarini birlashtirishda davom etmoqda. 🌟 Dunyo bo‘ylab kelib tushgan 350 dan ortiq ariza orasidan 20 ta ozarbayjonlik startap o‘z innovatsion g‘oyalarini taqdim etdi. Natijada ikki finalchi — “Starkids” va “Tryverse” — joriy yil dekabr oyida Toshkentda bo‘lib o‘tadigan Grand Final bosqichiga yo‘llanma oldi. 🚀 Jarayon tobora qizg‘in tus olmoqda — yakuniy bosqichga yaqinlashar ekanmiz, yangiliklarni kuzatishda davom eting! 📖To‘liq maqolani o'qing: startupbase.uz — 🇦🇿Azerbaijan Hosts Its Best Startup Project Semifinal! The Digital Startup Awards (DSA) continues to bring together the brightest founders from across the region. 🌟 Out of 350+ applications globally, 20 startups from Azerbaijan took the stage to pitch their innovative ideas — and two finalists, Starkids and Tryverse, have earned their spot in the Grand Final in Tashkent this December. 🚀 The momentum is growing — stay tuned as we move closer to the final showdown! 📖Read the full article here: startupbase.uz #DSA#BestStartupProject#ITParkUzbekistan#AzerbaijanStartups#CentralAsiaStartups

🇦🇿 Ozarbayjonda Best Startup Project yarim final bosqichi bo‘lib o‘tdi! “Digital Startup Awards” (DSA) butun mintaqa bo‘ylab eng iqtidorli startap asoschilarini birlashtirishda davom etmoqda. 🌟 Dunyo bo‘ylab kelib tushgan 350 dan ortiq ariza orasidan 20 ta ozarbayjonlik startap o‘z innovatsion g‘oyalarini taqdim etdi. Natijada ikki finalchi — “Starkids” va “Tryverse” — joriy yil dekabr oyida Toshkentda bo‘lib o‘tadigan Grand Final bosqichiga yo‘llanma oldi. 🚀 Jarayon tobora qizg‘in tus olmoqda — yakuniy bosqichga yaqinlashar ekanmiz, yangiliklarni kuzatishda davom eting! 📖To‘liq maqolani o'qing: startupbase.uz — 🇦🇿Azerbaijan Hosts Its Best Startup Project Semifinal! The Digital Startup Awards (DSA) continues to bring together the brightest founders from across the region. 🌟 Out of 350+ applications globally, 20 startups from Azerbaijan took the stage to pitch their innovative ideas — and two finalists, Starkids and Tryverse, have earned their spot in the Grand Final in Tashkent this December. 🚀 The momentum is growing — stay tuned as we move closer to the final showdown! 📖Read the full article here: startupbase.uz #DSA#BestStartupProject#ITParkUzbekistan#AzerbaijanStartups#CentralAsiaStartups

DeepSeek-V4 预览版发布并开源：1M 上下文成标配,Agent 能力达开源最佳 4 月 24 日,DeepSeek 正式上线并同步开源 DeepSeek-V4 系列预览版,按规模分为 V4-Pro 和 V4-Flash 两款,全系标配 1M(一百万)token 超长上下文,在 Agent 能力、世界知识与推理性能上均达到国内与开源领域领先水平。官网 chat.deepseek.com、App 与 API 已同步更新,API 通过 model_name=deepseek-v4-pro 或 deepseek-v4-flash 调用。 📌 V4-Pro:对齐顶级闭源模型 - Agent 能力大幅提升:Agentic Coding 评测达开源模型最佳,已成为 DeepSeek 公司内部员工使用的编程模型 - 编程实测:使用体验优于 Claude Sonnet 4.5,交付质量接近 Opus 4.6 非思考模式,与 Opus 4.6 思考模式仍有一定差距 - 世界知识:在开源模型中大幅领先,仅稍逊于顶尖闭源 Gemini-Pro-3.1 - 推理性能:数学、STEM、竞赛代码测评超越所有已公开评测的开源模型,比肩世界顶级闭源 📌 V4-Flash:经济高效的小弟 - 参数和激活更小,API 服务更快、更便宜 - 推理能力接近 V4-Pro,世界知识储备稍逊 - Agent 评测:简单任务与 V4-Pro 旗鼓相当,高难度任务仍有差距 ⚙️ 结构创新:全新注意力 + DSA DeepSeek-V4 开创全新的 token 维度压缩注意力机制,结合 DeepSeek Sparse Attention(DSA 稀疏注意力),实现全球领先的长上下文能力,相比传统方法大幅降低计算和显存需求。官方明确:1M 上下文从此成为 DeepSeek 所有官方服务的标配。 🛠 Agent 生态专项适配 针对 Claude Code、OpenClaw、OpenCode、CodeBuddy 等主流 Agent 产品做了适配优化,代码任务、文档生成、PPT 生成等场景均有提升。 📎 API 细节 - base_url 不变,model 改为 deepseek-v4-pro 或 deepseek-v4-flash - 同时支持 OpenAI ChatCompletions 与 Anthropic 接口 - 均支持思考/非思考模式,思考模式可通过 reasoning_effort 参数设置强度(high/max),复杂 Agent 场景建议使用 max - 旧模型名 deepseek-chat(对应 V4-Flash 非思考)与 deepseek-reasoner(对应 V4-Flash 思考)将于 2026-07-24 停用 🔙 背景 - 2025-12-01 DeepSeek 发布 V3.2 正式版与 V3.2-Speciale,首次引入 DSA 稀疏注意力与「思考融入工具调用」,Speciale 拿下 IMO/CMO/IOI/ICPC 2025 金牌,推理能力对齐 Gemini-3.0-Pro - 此后 DeepSeek 进入约 5 个月的大版本静默期,社区一度质疑团队「是否还在」;V4 的发布以百万上下文普惠 + 开源权重的方式正面回应 - V3.2 → V4 演进主线清晰:从 MoE+MLA → DSA 稀疏注意力 → token 维度压缩新注意力,长上下文效率持续压榨 ⚔️ 竞品对比 - vs Claude Opus 4.6:编程交付质量已接近 Opus 4.6 非思考模式,思考模式仍有差距 - vs Claude Sonnet 4.5:Agentic Coding 使用体验反超 - vs Gemini-Pro-3.1:世界知识稍逊,推理已比肩 - vs 国产阵营:紧接智谱 GLM-5.1(SWE-Bench Pro 全球第一)、千问 Qwen3.6-Plus(Terminal-Bench 超 Opus 4.5)、MiniMax M2.7、腾讯混元 Hy3 preview、小米 MiMo-V2.5 的密集发布节奏,国产开源模型在 2026 年 4 月形成集体冲顶态势 🏢 DeepSeek 近况 DeepSeek 自 R1 后长期以「低调迭代 + 开源开放」路线前行,V3.2 验证了 DSA 的工程可行性,V4 则进一步将其规模化推向百万上下文普惠。官方在后记中重申「长期主义」与 AGI 目标,不回应外界关于 R2 的猜测。 📎 资源链接 - 开源权重(HuggingFace):https://huggingface.co/collections/deepseek-ai/deepseek-v4 - 开源权重(ModelScope):https://modelscope.cn/collections/deepseek-ai/DeepSeek-V4 - 技术报告 PDF:https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/main/DeepSeek_V4.pdf - API 思考模式文档:https://api-docs.deepseek.com/zh-cn/guides/thinking_mode #DeepSeek#DeepSeekV4#AI大模型#开源模型#AIAgent#长上下文#DSA#AgenticCoding

🇲🇳 Mongoliyada Best Startup Project yarim final bosqichi boshlandi! “Digital Startup Awards” (DSA) dasturi qizg‘in davom etmoqda — Mongoliya rasman o‘zining ilk “Best Startup Project” tanlovini o‘tkazib, eng yaxshi finalchilarni saralab oldi! 🚀 350 dan ortiq arizalar orasidan UWISH va EGUNE Mo‘g‘uliston yarim finalida g‘olib chiqdi va joriy yilning dekabr oyida Toshkentda bo‘lib o‘tadigan Grand finalda ishtirok etadi. 📖To‘liq maqolani o'qing: startupbase.uz — 🇲🇳Mongolia Kicks Off the Best Startup Project Semifinals! The Digital Startup Awards (DSA) program is in full swing — and Mongolia has officially hosted the first Best Startup Project competition, selecting its top finalists! 🚀 Out of 350+ applications, UWISH and EGUNE became the winners of the Mongolian semifinal and will compete in the Grand Final in Tashkent this December. 📖Read the full article here: startupbase.uz #DSA#BestStartupProject#ITParkUzbekistan#StartupEcosystem#MongoliaStartups#Innovation#Entrepreneurship#CentralAsiaStartups

DeepSeek-V4 预览版发布并开源：1M 上下文成标配,Agent 能力达开源最佳 4 月 24 日,DeepSeek 正式上线并同步开源 DeepSeek-V4 系列预览版,按规模分为 V4-Pro 和 V4-Flash 两款,全系标配 1M(一百万)token 超长上下文,在 Agent 能力、世界知识与推理性能上均达到国内与开源领域领先水平。官网 chat.deepseek.com、App 与 API 已同步更新,API 通过 model_name=deepseek-v4-pro 或 deepseek-v4-flash 调用。 📌 V4-Pro:对齐顶级闭源模型 - Agent 能力大幅提升:Agentic Coding 评测达开源模型最佳,已成为 DeepSeek 公司内部员工使用的编程模型 - 编程实测:使用体验优于 Claude Sonnet 4.5,交付质量接近 Opus 4.6 非思考模式,与 Opus 4.6 思考模式仍有一定差距 - 世界知识:在开源模型中大幅领先,仅稍逊于顶尖闭源 Gemini-Pro-3.1 - 推理性能:数学、STEM、竞赛代码测评超越所有已公开评测的开源模型,比肩世界顶级闭源 📌 V4-Flash:经济高效的小弟 - 参数和激活更小,API 服务更快、更便宜 - 推理能力接近 V4-Pro,世界知识储备稍逊 - Agent 评测:简单任务与 V4-Pro 旗鼓相当,高难度任务仍有差距 ⚙️ 结构创新:全新注意力 + DSA DeepSeek-V4 开创全新的 token 维度压缩注意力机制,结合 DeepSeek Sparse Attention(DSA 稀疏注意力),实现全球领先的长上下文能力,相比传统方法大幅降低计算和显存需求。官方明确:1M 上下文从此成为 DeepSeek 所有官方服务的标配。 🛠 Agent 生态专项适配 针对 Claude Code、OpenClaw、OpenCode、CodeBuddy 等主流 Agent 产品做了适配优化,代码任务、文档生成、PPT 生成等场景均有提升。 📎 API 细节 - base_url 不变,model 改为 deepseek-v4-pro 或 deepseek-v4-flash - 同时支持 OpenAI ChatCompletions 与 Anthropic 接口 - 均支持思考/非思考模式,思考模式可通过 reasoning_effort 参数设置强度(high/max),复杂 Agent 场景建议使用 max - 旧模型名 deepseek-chat(对应 V4-Flash 非思考)与 deepseek-reasoner(对应 V4-Flash 思考)将于 2026-07-24 停用 🔙 背景 - 2025-12-01 DeepSeek 发布 V3.2 正式版与 V3.2-Speciale,首次引入 DSA 稀疏注意力与「思考融入工具调用」,Speciale 拿下 IMO/CMO/IOI/ICPC 2025 金牌,推理能力对齐 Gemini-3.0-Pro - 此后 DeepSeek 进入约 5 个月的大版本静默期,社区一度质疑团队「是否还在」;V4 的发布以百万上下文普惠 + 开源权重的方式正面回应 - V3.2 → V4 演进主线清晰:从 MoE+MLA → DSA 稀疏注意力 → token 维度压缩新注意力,长上下文效率持续压榨 ⚔️ 竞品对比 - vs Claude Opus 4.6:编程交付质量已接近 Opus 4.6 非思考模式,思考模式仍有差距 - vs Claude Sonnet 4.5:Agentic Coding 使用体验反超 - vs Gemini-Pro-3.1:世界知识稍逊,推理已比肩 - vs 国产阵营:紧接智谱 GLM-5.1(SWE-Bench Pro 全球第一)、千问 Qwen3.6-Plus(Terminal-Bench 超 Opus 4.5)、MiniMax M2.7、腾讯混元 Hy3 preview、小米 MiMo-V2.5 的密集发布节奏,国产开源模型在 2026 年 4 月形成集体冲顶态势 🏢 DeepSeek 近况 DeepSeek 自 R1 后长期以「低调迭代 + 开源开放」路线前行,V3.2 验证了 DSA 的工程可行性,V4 则进一步将其规模化推向百万上下文普惠。官方在后记中重申「长期主义」与 AGI 目标,不回应外界关于 R2 的猜测。 📎 资源链接 - 开源权重(HuggingFace):https://huggingface.co/collections/deepseek-ai/deepseek-v4 - 开源权重(ModelScope):https://modelscope.cn/collections/deepseek-ai/DeepSeek-V4 - 技术报告 PDF:https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/main/DeepSeek_V4.pdf - API 思考模式文档:https://api-docs.deepseek.com/zh-cn/guides/thinking_mode #DeepSeek#DeepSeekV4#AI大模型#开源模型#AIAgent#长上下文#DSA#AgenticCoding

🚀 AI TRENDS | OpenAI to Be Regulated Under EU's Digital Services Act, Sources Say OpenAI is set to be regulated under the European Union's Digital Services Act (DSA), according to sources cited by Handelsblatt. The company will be classified as a very large online search engine under the new regulations. According to Jin10, this classification will subject OpenAI to stricter compliance requirements aimed at ensuring transparency and accountability in its operations. The DSA is part of the EU's broader effort to regulate digital services and platforms, enhancing user protection and addressing issues such as misinformation and data privacy. OpenAI's inclusion under this regulatory framework highlights the growing scrutiny of AI technologies and their impact on society. #AI#OpenAI#EU#DigitalServicesAct#DSA#Regulation#TechPolicy#DataPrivacy#Misinformation#Transparency#Accountability