\u65e5\u671f\uff1a2026-05-02<\/p>\n
\u672c\u671f\u805a\u7126\uff1a\u91cd\u70b9\u5173\u6ce8\u6a21\u578b\u53d1\u5e03\u4e0e release notes\u3001\u5b98\u65b9 engineering blog\u3001AI coding \/ agent \/ SRE\u3001\u8bc4\u6d4b\u699c\u5355\u53d8\u5316\u3001\u5f00\u53d1\u8005\u5b9e\u8df5\u535a\u5ba2\u3001\u6846\u67b6\u751f\u6001\u3001\u5f00\u6e90\u6a21\u578b\u4e0e\u771f\u5b9e\u7528\u6237\u89c6\u89d2\uff1b\u5f53 HN\u3001Reddit\u3001Hugging Face \u7b49\u793e\u533a\u6e90\u53ef\u8bbf\u95ee\u65f6\u4f18\u5148\u7eb3\u5165\u3002<\/p>\n
Artificial Analysis \u6700\u65b0\u6a21\u578b\u6392\u540d\u89c2\u5bdf<\/strong>\uff08Artificial Analysis\uff09<\/p>\n \u4e2d\u6587\u6458\u8981\uff1a<\/strong>Artificial Analysis \u6a21\u578b\u6392\u884c\u699c\u663e\u793a\uff0cGPT-5.5\uff08xhigh\/high\uff09\u4f4d\u5217\u667a\u80fd\u6307\u6570\u699c\u9996\uff0cClaude Opus 4.7\uff08max\uff09\u4e0e Gemini 3.1 Pro Preview \u7d27\u968f\u5176\u540e\u3002\u8f93\u51fa\u901f\u5ea6\u65b9\u9762\uff0cMercury 2 \u4ee5 859 tokens\/s \u9886\u5148\uff0cGranite 4.0 H Small \u8fbe 407 tokens\/s\u3002\u5ef6\u8fdf\u6700\u4f4e\u7684\u662f NVIDIA Nemotron 3 Nano\uff080.40 \u79d2\uff09\u4e0e Ministral 3 3B\uff080.47 \u79d2\uff09\u3002\u4ef7\u683c\u7aef\uff0cQwen3.5 0.8B \u4ee5\u6bcf\u767e\u4e07 tokens $0.02 \u6210\u4e3a\u6700\u4fbf\u5b9c\u6a21\u578b\u3002\u4e0a\u4e0b\u6587\u7a97\u53e3\u65b9\u9762\uff0cLlama 4 Scout \u652f\u6301 1000 \u4e07 tokens\uff0cGrok 4.1 Fast \u652f\u6301 200 \u4e07 tokens\u3002\u5e73\u53f0\u63d0\u4f9b\u667a\u80fd\u3001\u901f\u5ea6\u3001\u4ef7\u683c\u3001\u5ef6\u8fdf\u3001\u4e0a\u4e0b\u6587\u7b49\u591a\u7ef4\u5ea6\u5bf9\u6bd4\uff0c\u5e76\u533a\u5206\u5f00\u6e90\u4e0e\u95ed\u6e90\u6a21\u578b\u3002<\/p>\n English Summary:<\/strong> Artificial Analysis model rankings show GPT-5.5 (xhigh\/high) leading the Intelligence Index, followed by Claude Opus 4.7 (max) and Gemini 3.1 Pro Preview. For output speed, Mercury 2 leads at 859 tokens\/s with Granite 4.0 H Small at 407 tokens\/s. Lowest latency goes to NVIDIA Nemotron 3 Nano (0.40s) and Ministral 3 3B (0.47s). Qwen3.5 0.8B is cheapest at $0.02 per million tokens. Context window leaders are Llama 4 Scout (10M tokens) and Grok 4.1 Fast (2M tokens). The platform offers multi-dimensional comparisons across intelligence, speed, price, latency, and context window, distinguishing open weights from proprietary models.<\/p>\n \u539f\u6587\u94fe\u63a5<\/a><\/p>\n<\/li>\n Introducing Claude Opus 4.7<\/strong>\uff08Anthropic News\uff09<\/p>\n \u4e2d\u6587\u6458\u8981\uff1a<\/strong>Anthropic \u53d1\u5e03 Claude Opus 4.7\uff0c\u5728\u9ad8\u7ea7\u8f6f\u4ef6\u5de5\u7a0b\u4efb\u52a1\u4e0a\u8f83 Opus 4.6 \u663e\u8457\u63d0\u5347\uff0c\u5c24\u5176\u5728\u590d\u6742\u957f\u5468\u671f\u4efb\u52a1\u4e2d\u8868\u73b0\u51fa\u66f4\u5f3a\u7684\u4e25\u8c28\u6027\u4e0e\u4e00\u81f4\u6027\u3002\u6a21\u578b\u5177\u5907\u66f4\u9ad8\u5206\u8fa8\u7387\u7684\u89c6\u89c9\u80fd\u529b\uff0c\u5728\u4e13\u4e1a\u4efb\u52a1\u4e2d\u66f4\u5177\u5ba1\u7f8e\u4e0e\u521b\u610f\u3002\u5c3d\u7ba1\u6574\u4f53\u80fd\u529b\u4e0d\u53ca Claude Mythos Preview\uff0c\u4f46\u5728\u591a\u9879\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u8d85\u8d8a Opus 4.6\u3002\u4f5c\u4e3a\u9996\u6b3e\u90e8\u7f72\u7f51\u7edc\u5b89\u5168\u9632\u62a4\u7684\u6a21\u578b\uff0cOpus 4.7 \u5185\u7f6e\u81ea\u52a8\u68c0\u6d4b\u5e76\u963b\u65ad\u9ad8\u98ce\u9669\u7f51\u7edc\u5b89\u5168\u8bf7\u6c42\u7684 safeguard\uff0c\u5b89\u5168\u4e13\u4e1a\u4eba\u58eb\u53ef\u7533\u8bf7\u52a0\u5165 Cyber Verification Program \u8fdb\u884c\u5408\u6cd5\u5b89\u5168\u7814\u7a76\u3002\u6a21\u578b\u5df2\u5728 Claude \u4ea7\u54c1\u3001API\u3001Amazon Bedrock\u3001Google Cloud Vertex AI \u53ca Microsoft Foundry \u4e0a\u7ebf\uff0c\u5b9a\u4ef7\u7ef4\u6301 $5\/$25 \u6bcf\u767e\u4e07\u8f93\u5165\/\u8f93\u51fa tokens\u3002\u65e9\u671f\u6d4b\u8bd5\u8005\u53cd\u9988\u5176\u5728 CursorBench \u5f97\u5206\u4ece 58% \u63d0\u5347\u81f3 70%\uff0c\u5728 93 \u9879\u7f16\u7801\u57fa\u51c6\u4e0a\u89e3\u51b3\u7387\u63d0\u5347 13%\u3002<\/p>\n English Summary:<\/strong> Anthropic released Claude Opus 4.7, showing notable improvements over Opus 4.6 in advanced software engineering, particularly on difficult long-running tasks requiring rigor and consistency. The model features substantially better vision with higher resolution support and improved aesthetic creativity for professional tasks. While less broadly capable than Claude Mythos Preview, it outperforms Opus 4.6 across benchmarks. As the first model with cyber safeguards, Opus 4.7 automatically detects and blocks high-risk cybersecurity requests; security professionals can apply for the Cyber Verification Program for legitimate research. Available across Claude products, API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry at $5\/$25 per million input\/output tokens. Early testers report CursorBench scores jumping from 58% to 70%, with 13% improvement on a 93-task coding benchmark.<\/p>\n \u539f\u6587\u94fe\u63a5<\/a><\/p>\n<\/li>\n Featured An update on recent Claude Code quality reports<\/strong>\uff08Anthropic Engineering\uff09<\/p>\n \u4e2d\u6587\u6458\u8981\uff1a<\/strong>Anthropic \u53d1\u5e03\u6280\u672f\u590d\u76d8\uff0c\u89e3\u91ca\u8fd1\u671f Claude Code \u8d28\u91cf\u4e0b\u964d\u62a5\u544a\u7684\u6839\u672c\u539f\u56e0\uff0c\u6d89\u53ca\u4e09\u9879\u72ec\u7acb\u53d8\u66f4\uff1a3 \u6708 4 \u65e5\u5c06\u9ed8\u8ba4\u63a8\u7406 effort \u4ece high \u8c03\u81f3 medium \u4ee5\u964d\u4f4e\u5ef6\u8fdf\uff0c4 \u6708 7 \u65e5\u5df2\u56de\u6eda\uff1b3 \u6708 26 \u65e5\u9488\u5bf9\u95f2\u7f6e\u8d85 1 \u5c0f\u65f6\u4f1a\u8bdd\u7684\u7f13\u5b58\u4f18\u5316\u5b58\u5728 bug\uff0c\u5bfc\u81f4\u6bcf\u8f6e\u5bf9\u8bdd\u90fd\u6e05\u9664\u5386\u53f2\u63a8\u7406\u800c\u975e\u4ec5\u4e00\u6b21\uff0c\u9020\u6210\u6a21\u578b"\u5065\u5fd8"\u4e0e\u91cd\u590d\uff0c4 \u6708 10 \u65e5\u4fee\u590d\uff1b4 \u6708 16 \u65e5\u65b0\u589e\u51cf\u5c11\u5197\u957f\u8f93\u51fa\u7684\u7cfb\u7edf\u63d0\u793a\uff0c\u610f\u5916\u635f\u5bb3\u4ee3\u7801\u8d28\u91cf\uff0c4 \u6708 20 \u65e5\u56de\u6eda\u3002\u4e09\u9879\u95ee\u9898\u5747\u5df2\u89e3\u51b3\uff08v2.1.116\uff09\uff0cAPI \u672a\u53d7\u5f71\u54cd\u3002Anthropic \u5411\u6240\u6709\u8ba2\u9605\u8005\u91cd\u7f6e\u4f7f\u7528\u989d\u5ea6\uff0c\u5e76\u627f\u8bfa\u6539\u8fdb\u6d41\u7a0b\u4ee5\u9632\u6b62\u7c7b\u4f3c\u95ee\u9898\u3002\u6587\u7ae0\u8be6\u7ec6\u62ab\u9732\u4e86\u4ee3\u7801\u5ba1\u67e5\u3001\u6d4b\u8bd5\u4e0e\u5185\u90e8\u8bc4\u4f30\u672a\u80fd\u53ca\u65f6\u53d1\u73b0\u95ee\u9898\u7684\u6559\u8bad\u3002<\/p>\n English Summary:<\/strong> Anthropic published a postmortem explaining recent Claude Code quality degradation reports, identifying three separate changes: March 4 default reasoning effort changed from high to medium to reduce latency, reverted April 7; March 26 caching optimization for idle sessions over 1 hour had a bug causing thinking history to clear every turn instead of once, causing forgetfulness and repetition, fixed April 10; April 16 system prompt to reduce verbosity inadvertently hurt coding quality, reverted April 20. All issues resolved as of April 20 (v2.1.116), API unaffected. Anthropic reset usage limits for all subscribers and committed to process improvements. The post details how code reviews, testing, and internal evals failed to catch the issues initially.<\/p>\n \u539f\u6587\u94fe\u63a5<\/a><\/p>\n<\/li>\n Scaling Managed Agents: Decoupling the brain from the hands<\/strong>\uff08Anthropic Engineering\uff09<\/p>\n \u4e2d\u6587\u6458\u8981\uff1a<\/strong>Anthropic Engineering \u535a\u5ba2\u4ecb\u7ecd Managed Agents \u67b6\u6784\u8bbe\u8ba1\uff0c\u6838\u5fc3\u601d\u8def\u662f\u5c06 Agent \u7684"\u5927\u8111"\uff08harness \u4e0e\u6a21\u578b\uff09\u4e0e"\u53cc\u624b"\uff08sandbox \u4e0e\u5de5\u5177\u6267\u884c\uff09\u53ca"\u4f1a\u8bdd"\uff08\u4e8b\u4ef6\u65e5\u5fd7\uff09\u89e3\u8026\u3002\u901a\u8fc7\u865a\u62df\u5316\u62bd\u8c61\uff08session\u3001harness\u3001sandbox\uff09\uff0c\u5b9e\u73b0\u5404\u7ec4\u4ef6\u72ec\u7acb\u5931\u8d25\u4e0e\u66ff\u6362\uff0c\u907f\u514d\u4f20\u7edf\u5355\u4f53\u5bb9\u5668\u6210\u4e3a"\u5ba0\u7269"\uff08pet\uff09\u800c\u975e" cattle"\u3002\u5b89\u5168\u8bbe\u8ba1\u4e0a\uff0c\u51ed\u8bc1\u5b58\u50a8\u4e8e vault \u5916\uff0csandbox \u901a\u8fc7\u4ee3\u7406\u8c03\u7528 MCP \u5de5\u5177\uff0c\u786e\u4fdd Claude \u751f\u6210\u7684\u4ee3\u7801\u65e0\u6cd5\u89e6\u53ca\u654f\u611f\u4ee4\u724c\u3002\u4f1a\u8bdd\u65e5\u5fd7\u72ec\u7acb\u4e8e\u6a21\u578b\u4e0a\u4e0b\u6587\u7a97\u53e3\uff0c\u652f\u6301\u957f\u5468\u671f\u4efb\u52a1\u7684\u6301\u4e45\u5316\u4e0e\u6062\u590d\u3002\u8be5\u67b6\u6784\u4f7f Replit\u3001Notion\u3001Harvey \u7b49\u5ba2\u6237\u80fd\u591f\u6258\u7ba1\u957f\u5468\u671f Agent\uff0c\u540c\u65f6\u5141\u8bb8\u5e95\u5c42\u5b9e\u73b0\u81ea\u7531\u6f14\u8fdb\u800c\u4e0d\u7834\u574f\u63a5\u53e3\u5951\u7ea6\u3002<\/p>\n English Summary:<\/strong> Anthropic Engineering blog introduces Managed Agents architecture, decoupling the "brain" (harness and model) from the "hands" (sandbox and tool execution) and "session" (event log). Through virtualization abstractions (session, harness, sandbox), components can fail and be replaced independently, avoiding the "pet vs cattle" problem of monolithic containers. Security design stores credentials outside the sandbox in a vault, with MCP tools called via proxy so Claude-generated code cannot access sensitive tokens. Session logs persist independently from model context windows, enabling long-horizon task durability and recovery. This architecture allows customers like Replit, Notion, and Harvey to host long-running agents while permitting underlying implementations to evolve without breaking interface contracts.<\/p>\n \u539f\u6587\u94fe\u63a5<\/a><\/p>\n<\/li>\n Replit\u2019s Amjad Masad on the Cursor deal, fighting Apple, and why he\u2019d rather not sell<\/strong>\uff08TechCrunch AI\uff09<\/p>\n \u4e2d\u6587\u6458\u8981\uff1a<\/strong>Replit CEO Amjad Masad \u5728 TechCrunch StrictlyVC \u6d3b\u52a8\u4e0a\u8c08\u53ca Cursor \u88ab SpaceX \u4ee5 600 \u4ebf\u7f8e\u5143\u6536\u8d2d\u7684\u4f20\u95fb\uff0c\u8868\u793a Cursor \u8d1f 23% \u6bdb\u5229\u7387\u4f7f\u5176\u96be\u4ee5\u72ec\u7acb\u751f\u5b58\uff0c\u800c Replit \u5df2\u4fdd\u6301\u6b63\u6bdb\u5229\u7387\u8d85\u4e00\u5e74\uff0cARR \u6b63\u8fc8\u5411 10 \u4ebf\u7f8e\u5143\u3002Masad \u5f3a\u8c03\u5e0c\u671b\u4fdd\u6301\u72ec\u7acb\uff0c\u516c\u53f8\u6210\u7acb 10 \u5e74\u6765\u59cb\u7ec8\u81f4\u529b\u4e8e\u8ba9\u975e\u6280\u672f\u7528\u6237\u521b\u5efa\u8f6f\u4ef6\uff0c2024 \u5e74 9 \u6708\u63a8\u51fa\u7684 Agentic \u7f16\u7801\u4f53\u9a8c\u5f15\u9886\u4e86\u884c\u4e1a\u6f6e\u6d41\u3002\u4ed6\u8bc4\u4ef7 Anthropic \u5728\u6838\u5fc3 Agent \u5faa\u73af\u4e0e\u5de5\u5177\u8c03\u7528\u4e0a\u4ecd\u65e0\u654c\uff0cGPT-5 \u5feb\u901f\u8ffd\u8d76\uff0cGoogle Flash \u7cfb\u5217\u5728\u6027\u4ef7\u6bd4\u4e0a\u9886\u5148\u3002Replit \u7684\u51c0\u6536\u5165\u7559\u5b58\u7387\u9ad8\u8fbe 300%\uff0c\u5ba2\u6237\u4e00\u65e6\u91c7\u7528\u5168\u6808\u65b9\u6848\u901a\u5e38\u4e0d\u4f1a\u6d41\u5931\u3002Masad \u8fd8\u900f\u9732\u56e0 Replit \u652f\u6301\u751f\u6210 iOS \u5e94\u7528\uff0c\u82f9\u679c\u4ee5"\u4e0b\u67b6\u5a01\u80c1"\u4e3a\u7531\u5c01\u9501\u5176 App Store \u66f4\u65b0\u5df2\u8fbe\u6570\u6708\uff0c\u4ed6\u8003\u8651\u8bc9\u8bf8\u6cd5\u5f8b\u3002<\/p>\n English Summary:<\/strong> Replit CEO Amjad Masad discussed Cursor's reported $60 billion SpaceX acquisition at TechCrunch StrictlyVC, noting Cursor's negative 23% gross margins make independence difficult, while Replit has been gross margin positive for over a year with ARR approaching $1 billion. Masad emphasized desire to remain independent, having spent 10 years enabling non-technical users to build software and pioneering agentic coding in September 2024. He ranked Anthropic as undefeated on core agentic loops and tool calling, GPT-5 catching up quickly, and Google's Flash family leading on price-performance. Replit's net revenue retention reaches 300%, with low churn once customers adopt the full-stack solution. Masad revealed Apple has blocked Replit's App Store updates for months\u2014allegedly because Replit enables iOS app creation\u2014and he's considering legal action.<\/p>\n \u539f\u6587\u94fe\u63a5<\/a><\/p>\n<\/li>\n AWS Transform now automates BI migration to Amazon Quick in days<\/strong>\uff08AWS ML Blog\uff09<\/p>\n \u4e2d\u6587\u6458\u8981\uff1a<\/strong>AWS Transform \u63a8\u51fa BI \u8fc1\u79fb\u529f\u80fd\uff0c\u53ef\u5c06 Tableau \u548c Power BI \u4eea\u8868\u677f\u81ea\u52a8\u8fc1\u79fb\u81f3 Amazon QuickSight\u3002\u901a\u8fc7\u4e0e Wavicle Data Solutions \u5408\u4f5c\uff0c\u7528\u6237\u53ef\u5728 AWS Marketplace \u8ba2\u9605\u4e13\u7528 Agent\uff08Analyzer \u4e0e Converter\uff09\uff0c\u4ee5\u5bf9\u8bdd\u5f0f\u754c\u9762\u5b8c\u6210\u8fc1\u79fb\u3002Analyzer Agent \u8d1f\u8d23\u63d0\u53d6\u6e90 BI \u5143\u6570\u636e\u5e76\u751f\u6210\u517c\u5bb9\u6027\u8bc4\u4f30\u62a5\u544a\uff1bConverter Agent \u5219\u91cd\u5efa\u6570\u636e\u96c6\u3001\u8ba1\u7b97\u5b57\u6bb5\u3001\u53ef\u89c6\u5316\u56fe\u8868\u53ca\u53c2\u6570\u3002\u6574\u4e2a\u6d41\u7a0b\u5728\u5ba2\u6237 AWS \u8d26\u6237\u5185\u8fd0\u884c\uff0c\u6570\u636e\u4e0d\u51fa\u5883\uff0c\u652f\u6301\u5e76\u884c\u5904\u7406\u6570\u767e\u4e2a\u4eea\u8868\u677f\uff0c\u6709\u671b\u5c06\u8fc1\u79fb\u5468\u671f\u4ece\u6570\u6708\u7f29\u77ed\u81f3\u6570\u5929\u3002<\/p>\n English Summary:<\/strong> AWS Transform now automates BI migration to Amazon QuickSight, enabling customers to migrate Tableau and Power BI dashboards in days instead of months. Through AWS Marketplace, users subscribe to Wavicle's specialized agents\u2014Analyzer and Converter\u2014to perform a two-step, chat-based migration within their own AWS accounts. The Analyzer agent extracts metadata and generates compatibility assessments, while the Converter agent rebuilds datasets, calculated fields, visualizations, and parameters in QuickSight.<\/p>\n \u539f\u6587\u94fe\u63a5<\/a><\/p>\n<\/li>\n Meta Deploys Unified AI Agents to Automate Performance Optimization at Hyperscale<\/strong>\uff08InfoQ AI\/ML\uff09<\/p>\n \u4e2d\u6587\u6458\u8981\uff1a<\/strong>Meta \u53d1\u5e03\u57fa\u4e8e\u7edf\u4e00 AI Agent \u7684\u5bb9\u91cf\u6548\u7387\u5e73\u53f0\uff0c\u7528\u4e8e\u81ea\u52a8\u5316\u68c0\u6d4b\u548c\u89e3\u51b3\u5168\u7403\u57fa\u7840\u8bbe\u65bd\u4e2d\u7684\u6027\u80fd\u95ee\u9898\u3002\u8be5\u7cfb\u7edf\u7ed3\u5408\u5927\u8bed\u8a00\u6a21\u578b Agent\u3001\u7ed3\u6784\u5316\u5de5\u5177\u4e0e\u7f16\u7801\u5de5\u7a0b\u77e5\u8bc6\uff0c\u53ef\u6301\u7eed\u5206\u6790\u57fa\u7840\u8bbe\u65bd\u6027\u80fd\u3001\u8bc6\u522b\u4f4e\u6548\u73af\u8282\u5e76\u81ea\u52a8\u4f18\u5316\u3002Agent \u80fd\u591f\u67e5\u8be2\u6027\u80fd\u5206\u6790\u6570\u636e\u3001\u68c0\u67e5\u914d\u7f6e\u5e76\u5b9e\u65bd\u4f18\u5316\uff0c\u5c06\u8d44\u6df1\u5de5\u7a0b\u5e08\u7684\u4e13\u4e1a\u77e5\u8bc6\u8f6c\u5316\u4e3a\u53ef\u590d\u7528\u7684"\u6280\u80fd"\uff0c\u5b9e\u73b0\u8de8\u7ec4\u7ec7\u89c4\u6a21\u5316\u5e94\u7528\u3002\u8fd9\u6807\u5fd7\u7740\u4ece\u88ab\u52a8\u54cd\u5e94\u5f0f\u6027\u80fd\u7ba1\u7406\u5411\u6301\u7eed\u81ea\u52a8\u5316\u4f18\u5316\u7684\u8f6c\u53d8\uff0c\u4f7f\u5de5\u7a0b\u5e08\u5f97\u4ee5\u4e13\u6ce8\u4e8e\u9ad8\u4ef7\u503c\u5de5\u4f5c\uff0c\u540c\u65f6\u964d\u4f4e\u8d44\u6e90\u6d6a\u8d39\u548c\u529f\u8017\u3002<\/p>\n English Summary:<\/strong> Meta has deployed a unified AI agent platform for automated performance optimization at hyperscale. The system combines LLM-based agents with structured tooling and encoded engineering expertise to continuously detect and resolve infrastructure inefficiencies across Meta's global footprint. By operationalizing institutional knowledge into reusable agent "skills," the platform enables autonomous diagnosis and remediation of performance issues, reducing manual intervention and freeing engineers for higher-value work.<\/p>\n \u539f\u6587\u94fe\u63a5<\/a><\/p>\n<\/li>\n [AINews] Agents for Everything Else: Codex for Knowledge Work, Claude for Creative Work<\/strong>\uff08Latent Space\uff09<\/p>\n \u4e2d\u6587\u6458\u8981\uff1a<\/strong>OpenAI \u5927\u5e45\u6269\u5c55 Codex \u5b9a\u4f4d\uff0c\u4ece\u7f16\u7a0b\u52a9\u624b\u5347\u7ea7\u4e3a\u9762\u5411\u77e5\u8bc6\u5de5\u4f5c\u7684\u901a\u7528 Agent \u5e73\u53f0\u3002\u65b0\u7248\u672c Codex \u63a8\u51fa"Codex for Work"\u9875\u9762\uff0c\u660e\u786e\u7784\u51c6\u975e\u7f16\u7801\u573a\u666f\u5982\u6587\u6863\u5904\u7406\u3001\u7535\u5b50\u8868\u683c\u3001\u6f14\u793a\u6587\u7a3f\u548c\u51b3\u7b56\u8ffd\u8e2a\u3002\u529f\u80fd\u66f4\u65b0\u5305\u62ec\uff1aCUA \u901f\u5ea6\u63d0\u5347 42%\u3001\u54cd\u5e94\u5f0f\u6d4f\u89c8\u5668\u3001\/chronicle \u548c \/goal \u547d\u4ee4\uff0c\u4ee5\u53ca Microsoft\/Google\/Salesforce \u5957\u4ef6\u96c6\u6210\u3002\u4ea7\u54c1\u91c7\u7528\u52a8\u6001 UI \u8bbe\u8ba1\uff0c\u7531 Agent \u81ea\u4e3b\u51b3\u5b9a\u754c\u9762\u6d41\u7a0b\uff0c\u800c\u975e\u56fa\u5b9a\u6a21\u5f0f\u3002Sam Altman \u4e0e Greg Brockman \u5747\u5f3a\u8c03 Codex \u9002\u7528\u4e8e"\u4efb\u4f55\u8ba1\u7b97\u673a\u4efb\u52a1"\uff0c\u6807\u5fd7\u7740 OpenAI \u5c06\u7f16\u7801 Agent \u4ea7\u54c1\u5316\u4e3a\u901a\u7528\u8ba1\u7b97\u673a\u4f7f\u7528 Agent \u7684\u6218\u7565\u8f6c\u5411\u3002<\/p>\n English Summary:<\/strong> OpenAI has expanded Codex from a coding assistant into a general-purpose agent for knowledge work. The "Codex for Work" launch targets non-coding tasks including documents, spreadsheets, presentations, and research workflows. Key updates include 42% faster Computer Use, responsive browser capabilities, \/chronicle and \/goal commands, and integrations with Microsoft, Google, and Salesforce suites. The product features a dynamic UI that lets the agent route the interface experience rather than using fixed modes. With executives framing Codex as "for everyone, for any task done with a computer," OpenAI is signaling a strategic shift toward productizing computer-use agents beyond the developer niche.<\/p>\n \u539f\u6587\u94fe\u63a5<\/a><\/p>\n<\/li>\n GitHub Copilot CLI for Beginners: Interactive v. non-interactive mode<\/strong>\uff08GitHub AI\/ML\uff09<\/p>\n \u4e2d\u6587\u6458\u8981\uff1a<\/strong>GitHub \u53d1\u5e03 Copilot CLI \u521d\u5b66\u8005\u6307\u5357\uff0c\u8be6\u89e3\u4ea4\u4e92\u5f0f\u4e0e\u975e\u4ea4\u4e92\u5f0f\u4e24\u79cd\u6a21\u5f0f\u3002\u4ea4\u4e92\u6a21\u5f0f\u4e3a\u9ed8\u8ba4\u7684\u4f1a\u8bdd\u5f0f\u4f53\u9a8c\uff0c\u7528\u6237\u53ef\u4e0e Copilot \u8fdb\u884c\u591a\u8f6e\u5bf9\u8bdd\u3001\u8fed\u4ee3\u63d0\u95ee\uff0c\u9002\u5408\u63a2\u7d22\u6027\u6df1\u5ea6\u5de5\u4f5c\uff1b\u975e\u4ea4\u4e92\u6a21\u5f0f\u901a\u8fc7 `copilot -p` \u89e6\u53d1\uff0c\u9002\u7528\u4e8e\u5feb\u901f\u5355\u6b21\u67e5\u8be2\u5982\u4ee3\u7801\u7247\u6bb5\u751f\u6210\u6216\u4ed3\u5e93\u6458\u8981\uff0c\u6267\u884c\u540e\u7acb\u5373\u8fd4\u56de\u7ec8\u7aef\u3002\u6b64\u5916\uff0c\u7528\u6237\u53ef\u901a\u8fc7 `\/resume` \u6216 `–resume` \u547d\u4ee4\u6062\u590d\u4e4b\u524d\u7684\u4f1a\u8bdd\uff0c\u4fdd\u7559\u5b8c\u6574\u4e0a\u4e0b\u6587\u3002\u8be5\u7cfb\u5217\u8fd8\u9884\u544a\u540e\u7eed\u5c06\u6db5\u76d6\u659c\u6760\u547d\u4ee4\u4e0e MCP \u670d\u52a1\u5668\u7b49\u8fdb\u9636\u4e3b\u9898\u3002<\/p>\n English Summary:<\/strong> GitHub published a beginner's guide to Copilot CLI covering interactive and non-interactive modes. Interactive mode provides a chat-like session for iterative, exploratory work where users can ask follow-up questions and collaborate with Copilot. Non-interactive mode, accessed via `copilot -p`, delivers quick one-off answers for tasks like code snippets or repository summaries without entering a session. Users can resume previous sessions with `\/resume` or `–resume` to retain context.<\/p>\n \u539f\u6587\u94fe\u63a5<\/a><\/p>\n<\/li>\n [AINews] The Inference Inflection<\/strong>\uff08Latent Space\uff09<\/p>\n \u4e2d\u6587\u6458\u8981\uff1a<\/strong>AI \u884c\u4e1a\u6b63\u7ecf\u5386"\u63a8\u7406\u62d0\u70b9"\uff08Inference Inflection\uff09\uff0c\u8ba1\u7b97\u9700\u6c42\u4ece\u8bad\u7ec3\u8f6c\u5411\u63a8\u7406\u3002OpenAI CEO Sam Altman \u8868\u793a\u516c\u53f8\u5fc5\u987b\u6210\u4e3a"AI \u63a8\u7406\u516c\u53f8"\uff0cIntel CEO \u4e5f\u6307\u51fa CPU \u9700\u6c42\u6b63\u5728\u4e0a\u5347\u3002\u968f\u7740 AI Agent \u548c RL \u5de5\u4f5c\u8d1f\u8f7d\u589e\u957f\uff0c\u63a8\u7406\u8ba1\u7b97\u9700\u6c42\u5728\u4e24\u5e74\u5185\u589e\u957f\u7ea6 10,000 \u500d\uff0c\u53e0\u52a0 COVID \u671f\u95f4\u91c7\u8d2d\u7684 CPU \u8fdb\u5165\u66f4\u6362\u5468\u671f\uff0c\u53ef\u80fd\u5f15\u53d1 CPU \u4f9b\u5e94\u7d27\u5f20\u3002\u540c\u65f6\uff0cGPU \u63a8\u7406\u67b6\u6784\u4e5f\u5728\u53d8\u9769\uff1aPrefill\/Decode \u5206\u79bb\u6210\u4e3a\u5e38\u6001\uff0cNVIDIA\u3001Intel-Sambanova\u3001Amazon \u7b49\u7eb7\u7eb7\u63a8\u51fa\u76f8\u5e94\u65b9\u6848\u3002\u8fd9\u4e00\u8d8b\u52bf\u6807\u5fd7\u7740 AI \u57fa\u7840\u8bbe\u65bd\u4ece\u8bad\u7ec3\u5bc6\u96c6\u578b\u5411\u63a8\u7406\u5bc6\u96c6\u578b\u8f6c\u53d8\uff0c\u63a8\u7406\u8ba1\u7b97\u6b63\u6210\u4e3a\u6218\u7565\u6027\u8d44\u6e90\u3002<\/p>\n English Summary:<\/strong> The AI industry is experiencing an "Inference Inflection" as compute demand shifts from training to inference. Sam Altman stated OpenAI must become "an AI inference company," while Intel's CEO highlighted rising CPU demand for agent and RL workloads. Inference compute requirements have increased roughly 10,000x in two years, coinciding with the COVID-era CPU refresh cycle, potentially creating CPU shortages. GPU architectures are also evolving with prefill\/decode disaggregation becoming standard, as NVIDIA, Intel-Sambanova, and Amazon pursue similar approaches. This marks a fundamental shift from training-intensive to inference-intensive AI infrastructure, with inference compute emerging as a strategic resource.<\/p>\n \u539f\u6587\u94fe\u63a5<\/a><\/p>\n<\/li>\n Introducing Advanced Account Security<\/strong>\uff08OpenAI News\uff09<\/p>\n \u4e2d\u6587\u6458\u8981\uff1a<\/strong>OpenAI \u63a8\u51fa\u300c\u9ad8\u7ea7\u8d26\u6237\u5b89\u5168\u300d\u529f\u80fd\uff0c\u9762\u5411\u9ad8\u98ce\u9669\u7528\u6237\u53ca\u5b89\u5168\u654f\u611f\u4eba\u7fa4\u63d0\u4f9b\u589e\u5f3a\u4fdd\u62a4\u3002\u8be5\u529f\u80fd\u5f3a\u5236\u4f7f\u7528\u901a\u884c\u5bc6\u94a5\u6216\u7269\u7406\u5b89\u5168\u5bc6\u94a5\uff08\u5982 YubiKey\uff09\u767b\u5f55\uff0c\u7981\u7528\u5bc6\u7801\u548c\u77ed\u4fe1\/\u90ae\u4ef6\u6062\u590d\u65b9\u5f0f\uff0c\u6539\u7528\u5907\u4efd\u5bc6\u94a5\u548c\u6062\u590d\u7801\uff1b\u4f1a\u8bdd\u6709\u6548\u671f\u7f29\u77ed\u5e76\u652f\u6301\u6d3b\u52a8\u76d1\u63a7\uff1b\u540c\u65f6\u81ea\u52a8\u6392\u9664\u5bf9\u8bdd\u6570\u636e\u7528\u4e8e\u6a21\u578b\u8bad\u7ec3\u3002OpenAI \u4e0e Yubico \u5408\u4f5c\u63d0\u4f9b\u4f18\u60e0\u786c\u4ef6\u5957\u88c5\uff0c\u5e76\u5ba3\u5e03\u81ea 2026 \u5e74 6 \u6708 1 \u65e5\u8d77\uff0c\u53c2\u4e0e\u300cTrusted Access for Cyber\u300d\u8ba1\u5212\u7684\u7528\u6237\u5fc5\u987b\u542f\u7528\u8be5\u529f\u80fd\u3002<\/p>\n English Summary:<\/strong> OpenAI introduces Advanced Account Security, an opt-in feature for high-risk users requiring passkeys or physical security keys (e.g., YubiKey) for phishing-resistant login while disabling password and SMS\/email recovery. It shortens session duration, enables activity monitoring, and automatically excludes conversations from model training. OpenAI partnered with Yubico for discounted hardware bundles, and will mandate enrollment for Trusted Access for Cyber participants starting June 1, 2026.<\/p>\n \u539f\u6587\u94fe\u63a5<\/a><\/p>\n<\/li>\n Where the goblins came from<\/strong>\uff08OpenAI News\uff09<\/p>\n \u4e2d\u6587\u6458\u8981\uff1a<\/strong>OpenAI \u53d1\u5e03\u6280\u672f\u535a\u5ba2\u89e3\u91ca GPT-5 \u7cfb\u5217\u6a21\u578b\u4e2d\u300c\u54e5\u5e03\u6797\u300d\u7b49\u5947\u5e7b\u751f\u7269\u9690\u55bb\u6cdb\u6ee5\u7684\u6839\u56e0\u3002\u95ee\u9898\u6e90\u81ea\u300cNerdy\u300d\u4eba\u683c\u5b9a\u5236\u529f\u80fd\u7684\u5f3a\u5316\u5b66\u4e60\u5956\u52b1\u4fe1\u53f7\u2014\u2014\u8be5\u4eba\u683c\u504f\u597d\u4fcf\u76ae\u8bed\u8a00\u98ce\u683c\uff0c\u65e0\u610f\u4e2d\u9ad8\u5956\u52b1\u4e86\u542b\u751f\u7269\u9690\u55bb\u7684\u8f93\u51fa\uff0c\u5bfc\u81f4\u8be5\u8868\u8fbe\u4e60\u60ef\u901a\u8fc7\u76d1\u7763\u5fae\u8c03\u548c\u504f\u597d\u6570\u636e\u53cd\u9988\u6269\u6563\u81f3\u5176\u4ed6\u573a\u666f\u3002\u5c3d\u7ba1 3 \u6708\u5df2\u4e0b\u7ebf Nerdy \u4eba\u683c\u5e76\u6e05\u7406\u8bad\u7ec3\u6570\u636e\uff0cGPT-5.5 \u4ecd\u56e0\u8bad\u7ec3\u542f\u52a8\u8f83\u65e9\u800c\u6b8b\u7559\u6b64\u73b0\u8c61\uff0c\u56e2\u961f\u5df2\u5728 Codex \u4e2d\u901a\u8fc7\u5f00\u53d1\u8005\u63d0\u793a\u8bcd\u7f13\u89e3\u3002OpenAI \u5f3a\u8c03\u6b64\u6848\u4f8b\u5c55\u793a\u4e86\u5956\u52b1\u4fe1\u53f7\u5bf9\u6a21\u578b\u884c\u4e3a\u7684\u610f\u5916\u5f71\u54cd\u53ca\u5efa\u7acb\u884c\u4e3a\u5ba1\u8ba1\u80fd\u529b\u7684\u91cd\u8981\u6027\u3002<\/p>\n English Summary:<\/strong> OpenAI published a technical blog explaining why GPT-5 models increasingly used goblin\/gremlin metaphors. The root cause was a reinforcement learning reward signal for the "Nerdy" personality that inadvertently favored creature-laden playful language. This tic spread via supervised fine-tuning and preference data feedback loops. Although the Nerdy personality was retired in March and training data filtered, GPT-5.5 still exhibits the behavior due to earlier training start; Codex now mitigates it via developer prompt instructions. OpenAI highlights this as a case study in reward signal shaping and the need for behavioral auditing tools.<\/p>\n \u539f\u6587\u94fe\u63a5<\/a><\/p>\n<\/li>\n Reading today's open-closed performance gap<\/strong>\uff08Interconnects\uff09<\/p>\n \u4e2d\u6587\u6458\u8981\uff1a<\/strong>Interconnects \u535a\u5ba2\u5206\u6790\u5f53\u524d\u5f00\u6e90\u4e0e\u95ed\u6e90\u5927\u6a21\u578b\u6027\u80fd\u5dee\u8ddd\u7684\u590d\u6742\u6027\u3002\u4f5c\u8005\u6307\u51fa\uff0c\u5355\u4e00\u7efc\u5408\u8bc4\u5206\uff08\u5982 Artificial Analysis Intelligence Index\uff09\u63a9\u76d6\u4e86\u80fd\u529b\u5206\u5e03\u7684\u7ec6\u5fae\u5dee\u522b\uff1a\u95ed\u6e90\u524d\u6cbf\u5b9e\u9a8c\u5ba4\u5728\u4ee3\u7801\u4e0e\u7ec8\u7aef\u4ee3\u7406\u4efb\u52a1\u4e0a\u6295\u5165\u5de8\u8d44\uff0c\u800c\u5f00\u6e90\u6a21\u578b\uff08\u5c24\u5176\u662f\u4e2d\u56fd\u5b9e\u9a8c\u5ba4\uff09\u901a\u8fc7\u84b8\u998f\u4e0e\u5ef6\u8fdf\u91c7\u8d2d\u6570\u636e\/\u73af\u5883\u4fdd\u6301\u8ffd\u8d76\u3002\u7136\u800c\uff0c\u968f\u7740\u300c\u524d\u6cbf\u300d\u4efb\u52a1\u8f6c\u5411\u4f1a\u8ba1\u3001\u6cd5\u5f8b\u3001\u533b\u7597\u7b49\u9700\u8981\u6602\u8d35\u79c1\u6709\u6570\u636e\u4e0e\u9886\u57df\u5de5\u5177\u6574\u5408\u7684\u4e13\u4e1a\u77e5\u8bc6\u5de5\u4f5c\uff0c\u5f00\u6e90\u6a21\u578b\u5c06\u9762\u4e34\u66f4\u5927\u6311\u6218\u3002\u4f5c\u8005\u8ba4\u4e3a\uff0c\u57fa\u51c6\u6d4b\u8bd5\u4e0e\u771f\u5b9e\u6027\u80fd\u7684\u76f8\u5173\u6027\u6b63\u5728\u51cf\u5f31\uff0c\u95ed\u6e90\u5382\u5546\u9700\u4e0d\u65ad\u91cd\u65b0\u5b9a\u4e49\u300c\u524d\u6cbf\u300d\u4ee5\u7ef4\u6301\u5546\u4e1a\u4f18\u52bf\u3002<\/p>\n English Summary:<\/strong> Interconnects blog analyzes the nuanced open-vs-closed model performance gap. The author argues that composite benchmarks (e.g., Artificial Analysis Intelligence Index) obscure capability distributions: closed frontier labs dominate coding and terminal-agent tasks, while open models (especially Chinese labs) keep pace via distillation and discounted data\/environment purchases. As the "frontier" shifts to specialized knowledge work (accounting, law, healthcare) requiring expensive private data and domain-specific tool integrations, open models will struggle more. The author notes declining correlation between benchmarks and real-world performance, and that closed labs must continually redefine the frontier to sustain commercial advantage.<\/p>\n \u539f\u6587\u94fe\u63a5<\/a><\/p>\n<\/li>\n Building an emoji list generator with the GitHub Copilot CLI<\/strong>\uff08GitHub AI\/ML\uff09<\/p>\n \u4e2d\u6587\u6458\u8981\uff1a<\/strong>GitHub \u535a\u5ba2\u4ecb\u7ecd\u4f7f\u7528 GitHub Copilot CLI \u5f00\u53d1\u300cEmoji List Generator\u300d\u7684\u5b9e\u6218\u6848\u4f8b\u3002\u8be5\u5de5\u5177\u4e3a\u7ec8\u7aef\u5e94\u7528\uff0c\u53ef\u5c06\u7528\u6237\u8f93\u5165\u7684\u5217\u8868\u81ea\u52a8\u8f6c\u6362\u4e3a\u5e26\u76f8\u5173\u8868\u60c5\u7b26\u53f7\u7684 Markdown \u683c\u5f0f\u5e76\u590d\u5236\u5230\u526a\u8d34\u677f\u3002\u5f00\u53d1\u8fc7\u7a0b\u4e2d\u4f7f\u7528\u4e86 GitHub Copilot CLI \u7684 Plan \u6a21\u5f0f\uff08Claude Sonnet 4.6 \u751f\u6210\u8ba1\u5212\uff09\u3001Autopilot \u6a21\u5f0f\uff08Claude Opus 4.7 \u5b9e\u73b0\uff09\u3001\u591a\u6a21\u578b\u5de5\u4f5c\u6d41\u3001allow-all \u5de5\u5177\u6807\u5fd7\u53ca GitHub MCP \u670d\u52a1\u5668\u3002\u9879\u76ee\u91c7\u7528 @opentui\/core \u6784\u5efa\u7ec8\u7aef UI\u3001@github\/copilot-sdk \u63d0\u4f9b AI \u80fd\u529b\u3001clipboardy \u5904\u7406\u526a\u8d34\u677f\uff0c\u4ee3\u7801\u5df2\u5f00\u6e90\u3002<\/p>\n English Summary:<\/strong> GitHub blog showcases building an "Emoji List Generator" using the GitHub Copilot CLI. The terminal app converts user lists into emoji-enhanced Markdown and copies results to clipboard. The workflow leveraged Copilot CLI's Plan mode (Claude Sonnet 4.6), Autopilot mode (Claude Opus 4.7), multi-model orchestration, allow-all tools flag, and the GitHub MCP server. The stack includes @opentui\/core for terminal UI, @github\/copilot-sdk for AI, and clipboardy for clipboard access; the project is open-sourced.<\/p>\n