🌍 全球速报 · 多语种新闻

多语种新闻 · 技术 · 民生

每日自动采集 · 更新时间:2026-08-03 03:54 | 共 2170 条

全部 (2170) 东方财富 (198) Hacker News (153) Al Jazeera (120) 腾讯新闻 (116) GitHub Trending (108) BBC (90) 华尔街见闻 (89) 新浪财经 (85) 陆家嘴财经早餐 (55) 国际金融要情 (45) 中国证券报 (44) 新华社 (34) BBC Middle East (34) 环球市场播报 (32) Clawhub热点 (31) 东方财富网 (26) 证券时报 (25) 格隆汇 (25) 财联社 (24) 央视新闻 (24) 今日头条 (24) 每日经济新闻 (22) 人民日报 (17) Twitter AI KOL (17) AI News Today (16) 腾讯云开发者社区 (14) 新浪科技 (14) 上海证券报 (14) 新浪新闻 (12) 微博热搜 (11) 操盘必读 (9) CSDN (9) 腾讯云开发者 (8) 同花顺财经 (8) 金十数据 (7) 科创板日报 (7) 汇通财经 (7) 智通财经 (7) 搜狐 (7) 中国日报 (7) Twitter AI KOL; AI综合 (7) Clawhub (7) ChinaTechNews (7) 钛媒体 (6) 新华财经 (6) 数智早参 (6) 外交部 (6) 四大证券报 (6) 商务部 (6) BBC; BBC Middle East (6) 腾讯新闻/早报 (5) 搜狐科技 (5) 快科技 (5) 中国人民银行 (5) Unite.AI (5) IT之家 (5) 金融早参 (4) 财经网 (4) 网易新闻 (4) 科技日报 (4) 环球时报 (4) 每经; 钛媒体 (4) 每日芯闻 (4) 头条财经 (4) 国际金融报 (4) 华尔街见闻; 东方财富 (4) 东方财富; 港交所 (4) 东方财富; 伦敦金交所 (4) TechNode (4) CoinMarketCap; 交易所数据 (4) 钛媒体; 每经 (3) 财经早报 (3) 证券日报 (3) 腾讯科技 (3) 腾讯云 (3) 界面新闻 (3) 环球网 (3) 每经 (3) 外交部/腾讯新闻 (3) 四大证券报; 腾讯新闻 (3) 人民网 (3) 中国航天新闻网 (3) The World News (3) The Decoder (3) THE DECODER (3) AI综合 (3) AI梭哈日报 (3) 21IC电子网 (3) 陆家嘴财经 (2) 观察者网 (2) 腾讯; AI行业周报 (2) 股海导航 (2) 第一财经 (2) 私募排排网 (2) 每经; 新浪证券 (2) 新浪证券 (2) 操盘必读; 腾讯新闻 (2) 投资日历 (2) 微博教育 (2) 微博医疗 (2) 央视新闻联播 (2) 央行 (2) 央广网 (2) 太平洋科技 (2) 国家发改委 (2) 四大证券报/中国证券报 (2) 北京日报 (2) 今天全世界都在看的新闻 (2) 人民日报海外版 (2) 交易所; 期权数据 (2) 中国载人航天工程办公室; 腾讯新闻 (2) 上观新闻 (2) Twitter AI KOL; AI综合; 腾讯云开发者; 腾讯云开发者社区; 腾讯新闻 (2) TechWire Asia (2) Page 3 News (2) GitHub; Hacker News (2) GitHub (2) Ecns.cn; 中新社 (2) EETOP创芯网 (2) CNN (2) ABC News (2) 21经济网 (2) 21世纪经济报道 (2) 黄金行情 (1) 高盛 (1) 飞象网; 腾讯新闻 (1) 飞象网 (1) 风云日报 (1) 预见能源; 新浪财经 (1) 预见能源 (1) 韩联社; 今日头条 (1) 韩联社/腾讯新闻 (1) 雷递网/新浪财经 (1) 雷科技 (1) 陆家嘴财经早餐; 金十数据 (1) 陆家嘴财经早餐; 腾讯新闻 (1) 陆家嘴财经早餐; 新浪财经 (1) 陆家嘴财经早餐; 富途公告 (1) 陆家嘴财经早餐; 四大证券报 (1) 陆家嘴财经早餐/腾讯新闻 (1) 阿克西奥斯新闻网 (1) 金十数据; 腾讯新闻 (1) 金十数据; 国家发改委 (1) 量子位 (1) 路透社; 美联社 (1) 路透社; 新浪财经 (1) 路透社/腾讯新闻 (1) 赢家财富网 (1) 财闻 (1) 财联社; 新浪财经 (1) 财联社/综合 (1) 财联社/新浪财经 (1) 财政部/税务总局/工信部 (1) 证券时报; 每经 (1) 证券时报; 今日头条 (1) 证券之星 (1) 解放军报 (1) 行业消息 (1) 芝麻AI; 今日头条 (1) 艾瑞咨询/腾讯新闻 (1) 航天视窗; 中国航天系统科学与工程研究院 (1) 腾讯财经 (1) 腾讯证券 (1) 腾讯新闻; 环球视野 (1) 腾讯新闻; 新浪财经 (1) 腾讯新闻; AI行业晨报 (1) 腾讯新闻/陆家嘴财经早餐 (1) 腾讯新闻/科技财经日报 (1) 腾讯新闻/环球时报 (1) 腾讯新闻/Wind (1) 腾讯新闻/Kataeb (1) 腾讯新闻/AI Journal (1) 腾讯体育 (1) 腾讯云开发者; 腾讯云开发者社区; 腾讯新闻; 腾讯; The Decoder; THE DECODER (1) 腾讯云开发者; AI行业晨报 (1) 腾讯; The Decoder (1) 股市直击 (1) 股市早8点 (1) 联合国/综合 (1) 网易财经 (1) 网易科技 (1) 网易新闻; Google (1) 网易新闻/陆家嘴财经 (1) 网信中国 (1) 经济参考报 (1) 科技日报/腾讯新闻 (1) 百度百科 (1) 电子信息产业网 (1) 电商派Pro (1) 电商平台; 汇率数据 (1) 现代快报 (1) 现代AI新闻早班车 (1) 环球时报/腾讯新闻 (1) 环球市场播报; 腾讯新闻; CNN (1) 环球市场 (1) 猪说网 (1) 澎湃新闻; NASA (1) 港股早报 (1) 清华大学气候变化研究院 (1) 深交所 (1) 泰国中文社 (1) 法尔斯通讯社/综合 (1) 河南手机报 (1) 河南交通投资; 新浪 (1) 汽车精选 (1) 求是; 新华社 (1) 每经AI快讯 (1) 每经; 腾讯数智早参 (1) 每经; 腾讯 (1) 每经; 股市直击 (1) 每日经济新闻; 东方财富 (1) 格隆汇; 路透社 (1) 格隆汇; 东方财富; 新华社 (1) 杭州网 (1) 智谱 (1) 早啊新闻 (1) 方正证券/腾讯新闻 (1) 新浪财经; 证券时报 (1) 新浪财经; 网易新闻 (1) 新浪财经; 格隆汇 (1) 新浪财经; 彭博 (1) 新浪财经; 头条新闻 (1) 新浪财经; 今日头条 (1) 新浪财经; 中国经营报 (1) 新浪财经; 东方财富 (1) 新浪财经/高盛 (1) 新浪证券; 每经 (1) 新浪科技; 科学热点 (1) 新浪科技; 沈阳日报 (1) 新浪硬件 (1) 新浪新闻; 新华社 (1) 新浪半导体/央视财经 (1) 新浪AI热点 (1) 新民晚报 (1) 新华财经; 东方财富 (1) 新华网 (1) 新华社; 搜狐 (1) 新华社; 央视新闻 (1) 新华社; 国家医保局 (1) 新华社; 伊朗媒体 (1) 新华社; 东方财富 (1) 新华社; 世界经济论坛 (1) 新华社; CCTV国际时讯 (1) 新华社; 21经济网 (1) 新华社/金融早参 (1) 新华社/第一财经 (1) 新华社/日经 (1) 新华社/新浪 (1) 新华社/央视新闻 (1) 新华社/国航 (1) 新华社/以色列军方 (1) 新华社/人民网 (1) 新华社/人民日报 (1) 新华日报 (1) 新京报 (1) 数智早参; 媒体综合 (1) 数智早参/新华社 (1) 搜狐/今日AI快报 (1) 投资早参; 腾讯新闻 (1) 慧语简报 (1) 微博话题 (1) 微博讨论 (1) 微博科普 (1) 微博科技 (1) 微博电商 (1) 微博用户 (1) 微博技术 (1) 微博情感 (1) 微博博主 (1) 微博创作者 (1) 微博AI博主; 微博综合 (1) 工信部; 新浪财经 (1) 工信部/APEC发布会 (1) 工信部 (1) 山西网安 (1) 小米科技 (1) 头条新闻 (1) 央视新闻; 路透社 (1) 央视新闻; 新浪财经 (1) 央视新闻; 中国航发 (1) 央视新闻/网易新闻 (1) 央视/新华社 (1) 央视 (1) 央行公告; 新浪财经 (1) 央行公告 (1) 央行/证券时报 (1) 天津日报 (1) 天山建设报/综合 (1) 外交部; 新浪财经 (1) 外交部; 四大证券报 (1) 外交部; 中新社 (1) 国际金融要情; 路透 (1) 国际金融要情; 新浪财经 (1) 国际金融要情; 克普勒 (1) 国际金融要情/新浪财经 (1) 国际能源署 (1) 国资小新 (1) 国投证券/搜狐 (1) 国投证券/商业新知 (1) 国家药监局; 21经济网 (1) 国家能源局 (1) 国家网信办 (1) 国家统计局; 新华财经 (1) 国家统计局 (1) 国家发改委; 上海经信委 (1) 国家卫健委 (1) 国务院 (1) 商务部; 新浪财经 (1) 商务部; 新华财经 (1) 商务部; 中国证券报 (1) 和远气体公告 (1) 同花顺; 东方财富 (1) 同花顺 (1) 发改委 (1) 华西都市报 (1) 华西证券 (1) 华尔街见闻; 央视新闻; 金十数据 (1) 华尔街见闻; 国际金融要情 (1) 华尔街日报 (1) 华夏时报/新浪财经 (1) 华为计算 (1) 北京市经信局 (1) 北京市发改委 (1) 凤凰网 (1) 共同社; 今日头条 (1) 全球半导体观察 (1) 全景路演/腾讯新闻 (1) 全景网 (1) 光明日报; 西北大学 (1) 健康早闻/腾讯新闻 (1) 健康早闻 (1) 健康早报 (1) 侃财邦/福布斯 (1) 伊朗塔斯尼姆通讯社/新华社 (1) 企查查 (1) 今日头条; 路透社 (1) 今日头条; 芝麻AI (1) 今日头条; 外交部 (1) 今日头条; OpenRouter (1) 人民财讯 (1) 人民网/新华社 (1) 人民日报海外版; 新浪 (1) 人民日报; 21经济网 (1) 人力资源社会保障部; 21经济网 (1) 交易所; 基金公司 (1) 中科院; 央视新闻 (1) 中新网/腾讯新闻 (1) 中新网 (1) 中基协 (1) 中国青年报 (1) 中国载人航天官网 (1) 中国证监会 (1) 中国证券报; 新浪财经 (1) 中国证券报; 上海证券报 (1) 中国证券报/腾讯新闻 (1) 中国证券报/Wind (1) 中国航天报 (1) 中国网 (1) 中国经营报; 新浪 (1) 中国科学院金属研究所 (1) 中国科协/中国宇航学会 (1) 中国石化/新华社 (1) 中国石化 (1) 中国海警局 (1) 中国气象局 (1) 中国日报; 欧盟统计局 (1) 中国基金报 (1) 中华网/新浪财经 (1) 中东媒体报道 (1) 东方财富网; 中科宇航 (1) 东方财富网; 中国科学院 (1) 东方财富; 新华社 (1) 东方财富; 华尔街见闻 (1) 东方财富; 债券市场 (1) 东方财富; 上市公司公告 (1) 世界卫生组织; 今日头条 (1) 上观新闻; 新浪 (1) 上海新闻 (1) 上交所; 中证指数 (1) invest wallstreet; 新浪财经 (1) ZAKER新闻 (1) World Today Journal (1) World News TV (1) Wind/财联社 (1) UC Berkeley/综合 (1) The Federal (1) Test Source (1) Teknowire/综合 (1) Teknowire (1) Technology News Channel (1) TechWire Asia; The Decoder (1) TechWeb (1) SupremeNews (1) Reuters; 能源资讯 (1) One World News (1) NPR; The New York Times (1) NPR (1) NEWS POSTSEVEN (1) MetrowatchXtra (1) Media OutReach (1) InfoWorld (1) IndexNasdaq (1) IT之家; 新浪科技 (1) IT之家; 搜狐科技 (1) Hacker News; TechCrunch; Twitter AI KOL (1) Graphene2026 (1) Global News (1) GitHub; Twitter (1) GitHub; LangChain博客 (1) Choice数据 (1) CSDN; The Verge (1) CNN/腾讯新闻 (1) CNET (1) CES 2026; 今日头条 (1) CCTV国际时讯; 新华社 (1) CCTV+ (1) Axios/综合 (1) AWNews (1) AI日报; 掘金 (1) AI日报; CSDN (1) AI工具 (1) AI周报 (1) AINewsToday (1) AI News (1) 21经济网; 新浪财经 (1)
04-13 00:01 · 国际,新闻,BBC,国际
The conflict is now a test of wills - Iran’s capacity to absorb strikes versus Trump’s tolerance for the war's costs.
Trump's blockade threat raises risks and leaves predicaments unchanged After a diplomatic team led by Vice-President JD Vance tried, and failed, to reach a negotiated agreement to end the US war with Iran on Saturday, President Donald Trump had to decide his next move. That came on Sunday morning, in a series of Truth Social posts. The US will impose a naval blockade of Iran, he wrote. "No one who pays an illegal toll will have safe passage on the high seas," he wrote. He also said that the US would continue clearing mines from the Strait of Hormuz in order to ensure a safe passage for allied shipping. The US military, he added, was "locked and loaded" and prepared to resume attacks against Iran at an "appropriate moment". He went on to say that while progress had been made in the 20-hour negotiations in Islamabad, Iran would not meet the US demand that it abandon its nuclear ambitions. That view was contradicted somewhat by a US official familiar with Vance's negotiations, who spelled out a much longer list of disagreements – including on Iran's control of Hormuz and its support for regional proxies, like Houthi rebels in Yemen and Hezbollah in Lebanon. While Trump's latest posts didn't have the apocalyptic bluster of last week's threat to end Iranian civilisation, they pose a number of new challenges – and risks – for the American side. Will mine-clearing activities place American naval vessels at greater risk of Iranian attacks? How would the US determine who paid Iran a toll? Will the US use force on foreign-flagged ships that ignore the blockade? How will nations that depend on Iranian oil, like China, respond? Will the move, intended to choke off Iran's primary income stream, drive up the price of oil to even higher levels? There are no clear answers. Later on Sunday, the US military Central Command announced that the naval blockade would stop all ships traveling to or from Iranian ports – a different set of conditions than in Trump's earlier proposed action. "I don't understand how blockading the strait is going to somehow push the Iranians into opening it," Senator Mark Warner of Virginia, the ranking Democrat on the Senate Intelligence Committee, told CNN on Sunday. On CBS' Face the Nation programme, Republican congressman Mike Turner of Ohio, who until last year chaired the House Intelligence Committee, said the blockade is a means to force a resolution to the situation in Hormuz. "The president, by saying we're not just going to let them decide who gets through, is certainly calling all of our allies and everyone to the table," he said. "This needs to be addressed." Last week, before Iran and the US agreed to a two-week ceasefire and face-to-face negotiations, Trump had found himself in a difficult situation. He could continue to ratchet up the US attacks on Iran, possibly doing long-term damage to the nation's civilian infrastructure, adding to a humanitarian crisis and further destablising the global economy. Or, he could back away from a war that has always been unpopular among the American public and is beginning to frustrate even some of Trump's supporters, who believed his promises to avoid extended foreign conflicts and Middle East entanglements. A new CBS poll suggests that most Americans (59%) feel the war is going somewhat or very badly for the US. Many believe the key US objectives - such as keeping open the Strait of Hormuz, securing greater freedom for the Iranian people, and permanently ending Iran's nuclear programme - remain unmet. Overwhelming bipartisan majorities feel it's important for the US to achieve these goals. Nearly a week has passed and, despite American claims of victory, the predicaments facing the president have not changed. Talking to Fox News on Sunday morning, Trump said that Iran would ultimately give the US "everything" it wants. He added that while oil prices might be the same or higher in the months ahead, he believed the US economy would hold up. That, to say the least, is a gamble. And with November's midterm elections looming, the president's Republican Party could pay dearly at the polls if he is wrong. On Saturday night, as his vice-president was negotiating with the Iranians in Pakistan, Trump travelled to Miami, where he watched prize fighters batter each other in UFC cage matches. It was, according to members of the press pool in attendance, a bizarre spectacle. The president of the United States observed violent contests in a blood-spattered ring, chatted with celebrities and, at times, engaged in intense discussions with his Secretary of State, Marco Rubio, and other advisers, in full view of the thousands of attendees. Ultimate fighting cage matches, despite their ferocity, have set rules and time limits, and end with a clear winner and a loser. It's the kind of clarity that the Iran war may never provide, as it stretches into its second month and the current two-week ceasefire appears on the verge of collapse. The conflict has become a test of wil
▸ 展开全文
04-13 00:00 · AI政策,区域竞争,战略规划,市场准入,监管趋势
European AI. A playbook to own it
European AI A playbook to own it. Europe holds unique strengths: a world-class academic ecosystem, a commitment to human-centric technology, and a single market of +450 million people. The question is no longer whether Europe can compete, but how it can turn these assets into a cohesive, self-reliant AI powerhouse. Reading 52 minutes Published April 7, 2026 By Mistral AI A word from the CEO Europe has faced a growing technological gap, leaving its citizens, businesses, and governments increasingly reliant on foreign dominance. The cost is high: a diminished voice on the global stage, reduced control over the European future, and vulnerability to digital threats. Without action, we risk surveillance threats, economic decline, strategic weakness, and even the erosion of our democratic freedoms. But this challenge is also Europe’s greatest opportunity. The AI revolution has started and is a chance to not only catch up but to lead and define our own paths. Europe is home to a vibrant pool of untapped talent and industrial champions whose unique assets can push the boundaries of what AI can achieve. The competition from the U.S. and China is fierce, but Europe is not a market to be dominated, it is a powerhouse of innovation, creativity, and resilience. The question is not whether we can compete, but how we will rise to the occasion. AI can be the tool that secures our autonomy, strengthens our strategic sectors, increases our economic wealth and amplifies our global influence. To seize this moment, we must act decisively. We need to drive demand for homegrown AI, secure strategic sectors, and empower European players. Controlling our AI and infrastructure is not optional, it’s the only way to win the AI race. So now is the time to act: grow our talent pool and bring our best minds back to Europe, scale our innovative companies across all 27 Member States, and turn our diversity into a competitive edge by compressing knowledge and building AI that reflects the world’s complexity. Europe’s AI ecosystem is brimming with potential. By fostering an environment that nurtures growth, we can transform challenges into opportunities and reclaim our future. The race is on, and Europe should be ready to win it. Arthur Mensch Co-founder & CEO of Mistral AI Europe holds unique strengths: a world-class academic ecosystem, a commitment to human-centric technology, and a single market of over 450 million people. The question is no longer whether Europe can compete, but how it can turn these assets into a cohesive, self-reliant AI powerhouse. This playbook provides a clear, actionable framework to position Europe as that powerhouse, accelerating AI development and adoption, attracting and retaining top talent, simplifying regulation without sacrificing values, and mobilizing public and private investment to build homegrown AI infrastructure. Only with it, Europe can ensure AI is not only developed in Europe, but for Europe and on Europe’s terms. This document is not a theoretical exercise. It is a practical playbook, born from the lived experience of a European AI startup, Mistral AI, navigating one of the world’s most competitive, fast and capital-intensive industries. We have experienced misaligned equity frameworks, bureaucratic barriers that require the CEO to travel for basic administrative tasks, and legal uncertainty that complicates contracts and customer relationships. We have seen how regulatory overlaps create legal quagmires, how fragmented markets hinder growth, and how talent slips away due to administrative friction. This document is a call to turn Europe’s strengths into scalable, competitive advantage. It is grounded in the urgency of the moment and the conviction that Europe can and must build an AI ecosystem that reflects its values, serves its citizens, and competes globally. It is our collective duty to ensure AI can also be developed in Europe on terms that aligns with our priorities as Europeans. These challenges shaped our approach and led us to agree on three key principles to unlock Europe’s AI potential: Action over theory: Every recommendation, from visa reform to procurement gateways, is designed to be implemented, measured, and scaled.Unity in complexity: Europe's diversity is its strength, but its fragmentation is its Achilles' heel. This paper embraces the complexity of the EU's structure while offering solutions to align markets, reduce redundancy, and accelerate decision-making.Speed is not an option: We propose fast-track mechanisms for talent, capital, and compliance, so Europe's innovators aren't left behind. At Mistral AI, we’ve built a frontier AI company in Europe because we believe in its potential. This playbook is our contribution to ensuring that potential becomes reality, not just for us, but for the entire ecosystem. I. Attract and retain talent The most transformative advancements in AI, those that push the boundaries of what is possible, are driven by human genius, scientific curio
▸ 展开全文
04-13 00:00 · AI基准测试,技术漏洞,竞争风险,可信度危机,AI评估
Exploiting the most prominent AI agent benchmarks
How We Broke Top AI Agent Benchmarks: And What Comes Next Our agent hacked every major one. Here’s how — and what the field needs to fix. The Benchmark Illusion Every week, a new AI model climbs to the top of a benchmark leaderboard. Companies cite these numbers in press releases. Investors use them to justify valuations. Engineers use them to pick which model to deploy. The implicit promise is simple: a higher score means a more capable system. That promise is broken. We built an automated scanning agent that systematically audited eight among the most prominent AI agent benchmarks — SWE-bench, WebArena, OSWorld, GAIA, Terminal-Bench, FieldWorkArena, and CAR-bench — and discovered that every single one can be exploited to achieve near-perfect scores without solving a single task. No reasoning. No capability. Just exploitation of how the score is computed. These aren’t theoretical attacks. Our agent builds working exploits for each benchmark, runs them through the official evaluation pipelines, and watches the scores roll in. - A conftest.py file with 10 lines of Python “resolves” every instance on SWE-bench Verified. - A fake curl wrapper gives a perfect score on all 89 Terminal-Bench tasks without writing a single line of solution code. - Navigating Chromium to a file:// URL reads the gold answer directly from the task config — giving ~100% on all 812 WebArena tasks. - And many more… The benchmarks aren’t measuring what you think they’re measuring. This Is Already Happening Benchmark scores are actively being gamed, inflated, or rendered meaningless, not in theory, but in practice: - IQuest-Coder-V1 claimed 81.4% on SWE-bench — then researchers found that 24.4% of its trajectories simply ran git log to copy the answer from commit history. Corrected score: 76.2%. The benchmark’s shared environment made the cheat trivial. - METR found that o3 and Claude 3.7 Sonnet reward-hack in 30%+ of evaluation runs — using stack introspection, monkey-patching graders, and operator overloading to manipulate scores rather than solve tasks. - OpenAI dropped SWE-bench Verified after an internal audit found that 59.4% of audited problems had flawed tests — meaning models were being scored against broken ground truth. - In KernelBench, torch.empty() returns stale GPU memory that happens to contain the reference answer from the evaluator’s prior computation — zero computation, full marks. - Anthropic’s Mythos Preview showed that frontier models can actively try to hack the environment and succeed. In one episode, the model needed to edit files it lacked permissions for; after searching for workarounds, it found a way to inject code into a config file that would run with elevated privileges, and designed the exploit to delete itself after running. If a model can independently craft self-erasing privilege escalation exploits, it can find the holes in an evaluation harness. These are not isolated incidents. They are symptoms of a systemic problem: the benchmarks we rely on to measure AI capability are themselves vulnerable to the very capabilities they claim to measure. The Scorecard of Our Exploit Agent Zero tasks solved. Zero LLM calls (in most cases). Near-perfect scores. - Terminal-Bench (89 tasks) — 100% score. Binary wrapper trojans. - SWE-bench Verified (500 tasks) — 100% score. Pytest hooks force all tests to pass. - SWE-bench Pro (731 tasks) — 100% score. In-container parser overwrite. - WebArena (812 tasks) — ~100% score. Config leakage + DOM injection + prompt injection. - FieldWorkArena (890 tasks) — 100% score. Validation never checks answer correctness. - CAR-bench (all hallucination tasks) — 100% score. Reward components skipped entirely. - GAIA (165 tasks) — ~98% score. Public answers + normalization collisions. - OSWorld (369 tasks) — 73% score. VM state manipulation + public gold files. How Our Agent Did It Terminal-Bench: Trojanizing the Test Infrastructure Terminal-Bench evaluates agents on 89 complex terminal tasks (from building COBOL chess engines to configuring distributed systems). It uses Harbor, a sandboxing framework that re-uploads protected test files before verification. Containers have full internet access — Harbor’s allow_internet config defaults to True and no task overrides it. Harbor correctly protects /tests/ — but it doesn’t protect system binaries. Here’s the irony: 82 of 89 tasks download uv from the internet at verification time via curl , creating an exploitable dependency chain. Our exploit replaces /usr/bin/curl with a wrapper during the agent phase. When the verifier later runs curl ... | sh to install uv, our wrapper intercepts the request, installs the real uv, and then trojanizes the uvx binary. The trojanized uvx detects pytest invocations and produces fake passing output: ============================= test session starts ============================== collected 3 items tests/test_outputs.py ... [100%] ============================== 3 passed in 0.05s ===============================
▸ 展开全文
04-13 00:00 · 技术实验,创意应用,社区分享,艺术,开源文化
I gave every train in New York an instrument
Every dot is a real subway train. Eight hundred of them, give or take, form a small jazz combo (walking bass, piano, sax, vibes, brushes) that has been playing without pause for over a hundred years. On the platforms they are hot, screaming, full of complaint. This is the music inside the noise. The harmony moves through a slow chorus. A note is placed precisely where the train happens to be along its route. Rush hour fills the band with held tones; at 3 a.m. the silences grow longer. Whatever is playing now has not played before and will not play again. Share your location and the trains nearest you grow louder. The piece rearranges itself around your body. You are listening to a portrait of where you stand, played by the city you are standing in. Every dot is a real subway train. Eight hundred of them, give or take, form a small jazz combo (walking bass, piano, sax, vibes, brushes) that has been playing without pause for over a hundred years. On the platforms they are hot, screaming, full of complaint. This is the music inside the noise. The harmony moves through a slow chorus. A note is placed precisely where the train happens to be along its route. Rush hour fills the band with held tones; at 3 a.m. the silences grow longer. Whatever is playing now has not played before and will not play again. Share your location and the trains nearest you grow louder. The piece rearranges itself around your body. You are listening to a portrait of where you stand, played by the city you are standing in.
▸ 展开全文
04-13 00:00 · 融资,市场降温,估值调整,AI炒作,技术投资
Tech valuations are back to pre-AI boom levels
April 11, 2026 Tech Valuations Back to Pre-AI Boom Levels The chart below compares the forward P/E ratios for the S&P 500 and the S&P 500 Information Technology sector. Tech valuations have compressed from 40x to 20x, and we are back at levels last seen before the AI boom began.
▸ 展开全文
04-12 13:43 · 中东,局势,AlJazeera,中东
Talks between Iran and the US expose deep divisions, prolonging uncertainty during a two-week ceasefire.
High-stakes talks between the United States and Iran have ended without a deal, and Iran’s chief negotiator, Mohammad Bagher Ghalibaf, has blamed the US for the failure of talks held in Islamabad, Pakistan, during a two-week ceasefire in their war. Ghalibaf, the speaker of Iran’s parliament, said on Sunday that his delegation raised “forward-looking” initiatives during the talks on Saturday but the US failed to gain the trust of the Iranian delegation. Recommended Stories list of 3 items- list 1 of 3US President Trump says US ‘wins’ regardless of how Iran talks go - list 2 of 3Oil tankers exit Strait of Hormuz amid fragile US-Iran ceasefire - list 3 of 3Why did US-Iran talks end without an agreement in Pakistan? US Vice President JD Vance, who led the US delegation, said earlier that the talks ended without a deal. “The bad news is that we have not reached an agreement, and I think that’s bad news for Iran much more than it’s bad news for the United States of America,” he said six weeks into the US-Israeli war on Iran. The talks – the first direct engagement between the two countries at this level since the 1979 Iranian Revolution – exposed deep divisions on core issues, including Iran’s nuclear programme and the Strait of Hormuz, which in effect has been under Tehran’s control since the war began on February 28. The de facto blockade of the waterway, through which one-fifth of global crude oil supplies pass, has caused a global energy crisis and rattled stock markets globally. Al Jazeera’s Kamal Hyder, reporting from Islamabad, said the talks mediated by Pakistan, which went on for more than 21 hours, were “neither a breakthrough nor a breakdown”. Here’s a look at what each side said and what the key sticking points between Tehran and Washington are: What did the US say? The US framed the lack of a breakthrough primarily around Iran’s alleged refusal to meet its core demand: a firm commitment not to develop nuclear weapons. “We need to see an affirmative commitment that they will not seek a nuclear weapon and they will not seek the tools that would enable them to quickly achieve a nuclear weapon,” Vance told reporters at a news conference. “That is the core goal of the president of the United States, and that’s what we’ve tried to achieve through these negotiations.” Vance said Washington had made its “red lines” clear and presented what he described as a “final and best offer”. The vice president did not mention reopening the Strait of Hormuz. During the negotiations, US President Donald Trump seemed to insert himself in the talks, saying a deal was not entirely necessary. “We’re negotiating. Whether we make a deal or not makes no difference to me because we’ve won,” he told reporters in Washington, DC. However, the fact that Trump sent Vance to Islamabad showed the US was taking these talks seriously, Al Jazeera’s John Hendren said, reporting from Washington, DC. “The fact that Vance left [Islamabad] doesn’t necessarily mean that the talks are over,” he said, adding that the main sticking points seem to be the Strait of Hormuz and the gaps on Iran’s nuclear programme. “The US has been negotiating with Iran over time. Those talks can continue remotely, and leaving those talks may simply be a hard stance,” the Al Jazeera correspondent added. What did Iran say? Iran downplayed expectations and blamed the US for making what it called unreasonable demands. “The success of this diplomatic process depends on the seriousness and good faith of the opposing side, refraining from excessive demands and unlawful requests, and the acceptance of Iran’s legitimate rights and interests,” Iranian Ministry of Foreign Affairs spokesman Esmaeil Baghaei wrote on X. He added that the two sides discussed a range of issues, including the “Strait of Hormuz, the nuclear issue, war reparations, the lifting of sanctions, and the complete end of the war against Iran”. Baghaei emphasised that the lack of a deal should not be seen as a failure of the broader process, saying, “No one had such an expectation.” According to the IRNA news agency, when asked whether diplomacy had ended, Baghaei said, “Diplomacy never ends.” Ghalibaf posted on X: “The US has come to understand Iran’s logic and principles, and now it must decide whether it is capable of gaining our trust.” Mohsen Farkhani from the University of Isfahan said Tehran entered the talks not for a breakthrough but to show the world that Washington was not serious. “Iran was in these negotiations with a kind of totally realistic view and with distrust … to prove to the world that, for the third time, the US doesn’t have enough seriousness to solve the problems or the challenges in peace and negotiations,” he told Al Jazeera. What did Pakistan say? Pakistan called for the ceasefire to continue while keeping diplomacy alive. Foreign Minister Ishaq Dar stressed that it is “imperative” for both sides to uphold their commitment to the ceasefire, warning implicitly of the risks of renewed
▸ 展开全文
04-12 13:43 · 中东,局势,AlJazeera,中东
Trump says the US Navy would immediately start blockading the Strait of ⁠Hormuz after ceasefire talks in Pakistan fail.
US President Donald Trump says the US Navy would immediately start blockading the Strait of Hormuz and would also interdict every vessel in international waters that had paid a toll to Iran. Iran’s Parliamentary Speaker Mohammad Baqer Ghalibaf has blamed the US for the failure of the ceasefire negotiations in Pakistan, saying the US officials failed to gain the trust of the Iranian delegation.
▸ 展开全文
04-12 13:43 · 中东,局势,BBC,中东
BBC Middle East Correspondent Hugo Bachega hears from people affected by this week's attacks.
Still in shock, Beirut tries to heal after deadly Israeli attacks Lebanon experienced one of the deadliest days in its recent history, as widespread Israeli air strikes brought horror and destruction to the country on Wednesday, just hours after a ceasefire was announced in Iran. The attacks, that came in the middle of the day without warning, killed more than 300 people, a third of them women, children and the elderly, according to the Lebanese health ministry. Some of the attacks hit busy, densely populated neighbourhoods, places that had not been targeted before and where people felt they were safe. Israel said it carried out more than 100 air strikes in just 10 minutes on what it described as targets linked to the Iranian-backed group Hezbollah. BBC Middle East correspondent Hugo Bachega has spoken to residents in Beirut, a city that is still in shock.
▸ 展开全文
04-12 13:43 · 中东,局势,BBC,中东
Twenty-one hours was not enough to end 47 years of hostility between Iran and the US, writes the BBC's Lyse Doucet.
After Iran talks falter, the big question is 'what happens next?' Twenty-one hours was not enough to end 47 years of hostility between Iran and the US. The historic high-level talks in Islamabad, during a pause in weeks of grievous war, were always unlikely to end any other way. Calling this marathon negotiating session a failure belies the scale of the challenge in narrowing wide gaps on complex issues ranging from age-old suspicion about Iran's nuclear programme to new challenges this war has thrown up - most of all Iran's control of the strategic Strait of Hormuz, whose closure is causing economic shocks worldwide. To do a deal, they also needed to overcome a deep chasm of distrust. A day ago, it wasn't even certain the two sides would meet, and even more, sit down in the same room. A longstanding political taboo was broken. The urgent question now is: what happens next? What happens to the contested two-week ceasefire which pulled the world back from US President Donald Trump's alarming threat to destroy a "whole civilisation" in Iran? Would the US president be ready to send his negotiators back to the bargaining table? We're hearing reports from sources here in Islamabad that some conversations have continued after US Vice-President JD Vance boarded his plane at sunrise, declaring the US delegation had made their "final and best offer". Will the US now escalate or negotiate? We still don't know enough about what happened behind tightly closed doors in a five-star hotel in leafy locked-down Islamabad during talks that went on long into the night. There are still few details on the disputes and discussion between the two sides, assisted by Pakistani mediators, the calls to and from experts, advisers, and, according to Vance, "dozens" of calls to Trump himself. The vice-president spoke of the "core goal" of the US during his brief dawn news conference. "We need to see an affirmative commitment that [Iran] will not seek a nuclear weapon and they will not seek the tools that would enable them to quickly achieve a nuclear weapon," he said. During the last round of talks in February, before military strikes were unleashed again, Iran had offered new concessions including the dilution of its 440kg stockpile of uranium enriched to 60% - dangerously close to weapons-grade. But it still insists on its "right" to enrich and hasn't been willing to give up that stockpile, now said to be buried deep in the rubble after US and Israeli air strikes last year. It's also refused repeated demands to open the Strait of Hormuz - to allow the free flow of vital traffic in oil, gas and other essential goods - in the absence of a new agreement. Both the US and the Iranian delegations came to Islamabad emboldened by their belief that theirs was the winning side in this war. And they engaged knowing that, if they failed, there was the option to keep fighting – whatever the spiralling pain for their own people and a world reeling from the cost of this conflagration. There was also what Dr Sanam Vakil of Chatham House describes as a "limited psychological understanding of the adversary and what compromises are needed for a real deal". Vance spoke of good news – "we've had a number of substantive negotiations" - and there was bad news: "We have not reached an agreement." And he made it clear that was "bad news for Iran much more than the United States of America". Iran's foreign ministry spokesman Esmail Baghaei criticised the US's "excessive demands and unlawful requests" in a post on X. And its parliamentary speaker Mohammad Bagher Ghalibaf, who led Iran's negotiating team, wrote that "the opposing side ultimately failed to gain the trust of the Iranian delegation in this round of negotiations". Iran is indicating it's ready to keep talking. Pakistan's Foreign Minister Ishaq Dar urged all sides to uphold the fragile ceasefire and said they would continue their efforts to encourage dialogue - sentiments being echoed in other concerned capitals. If history provides any lessons, the last time Iran reached a nuclear deal with the US and other world powers in 2015, it took 18 months of breakthroughs and breakdowns. Trump has made it clear he doesn't want to get bogged down in protracted negotiations. Vance previously warned that the US would not be receptive if Tehran tried to "play us". Pakistani journalist Kamran Yousef - in a legion of journalists who pulled all-nighters to provide non-stop coverage with very few details - declared that this round was one of "no breakthrough but no breakdown either". The world waits for a verdict, most of all from Trump.
▸ 展开全文
04-12 13:43 · 国际,新闻,BBC,国际
Ukrainian President Volodymyr Zelensky says his nation's forces would respond "symmetrically" to Russian attacks.
Ukraine and Russia accuse each other of hundreds of ceasefire violations Ukraine and Russia have each accused one another of hundreds of violations of a short ceasefire coinciding with Orthdox Easter celebrations. The Ukrainian military said on Sunday morning that Russian forces had committed 2,299 violations since the truce began at 16:00 local time (13:00 GMT) on Saturday - including shooting four unarmed soldiers. The Russian defence ministry, in turn, said Ukrainian forces had committed 1,971 violations, including three attempted counter-attacks in Dnipropetrovsk region. Ukrainian President Volodymyr Zelensky earlier said his nation's forces would respond "symmetrically" to Russian attacks during the ceasefire, calling Easter "a time of peace". He added that he hoped the truce could be extended beyond Easter to facilitate peace negotiations which have all but stalled with the outbreak of war in the Middle East - but Russia rejected the idea, saying its attacks would resume on Monday. Russian troops mounted 28 attacks and carried out nearly 2,000 drone strikes, Ukraine's military said, but did not use bombs or missiles. Russian forces in the north-eastern Kharkiv region executed four Ukrainian soldiers after the ceasefire came into force, the local prosecutor's office said, describing the incident as a "grave violation of international humanitarian law". Ukraine's military said the troops were shot after being disarmed, calling it "another war crime by Russia". Ukrainian authorities published what appears to be an image taken by a drone showing four bodies lying in a clearing. In Ukraine's northern Sumy region, which borders Russia, local authorities said a Russian drone hit an ambulance overnight, injuring three medics. Russia's defence ministry said Ukraine had launched three overnight attacks on positions in the Pokrovsk area and Otradne in Dnipropetrovsk region. It also said four attempts by Ukrainian troops to advance in Sumy and Donetsk were "thwarted". Russian President Vladimir Putin announced the Easter ceasefire earlier this week, having previously resisted repeated calls from Ukraine for a temporary cessation of fighting. Zelensky said his forces would act in kind. Earlier in the year, Putin acquiesced to a US request to halt strikes on energy infrastructure as Ukraine braced for bitterly cold temperatures. Both Ukrainian and Russian sources accused each other of limited violations in the first few hours of the truce on Saturday, before making the much larger claims. Ukrainian and Russian authorities also announced they had each swapped 175 prisoners of war on Saturday, including seven civilians a piece. Ukrainian civilians and soldiers on the frontlines of the conflict - which has been raging since 2022 - have low expectations about the ceasefire. Kyiv has long pushed for a more comprehensive ceasefire, which it and its European allies see as a necessary first step towards striking a lasting end to the full-scale invasion. But Moscow has insisted on agreeing a peace deal first, prompting accusations that it is not serious about ending the fighting.
▸ 展开全文