{ "date": "2026-10-04", "title": "秋粮收获与AI出游的执行落差", "url": "https://daily.yisiliu.io/20261004/", "summary": "全国秋粮收获已过三成，后期仍需防范稻飞虱；AI出游采访中的车程误差与营业时间错配，暴露了攻略落地时的问题。", "generated_at": "2026-10-04T07:46:15+08:00", "sections": [ { "id": "judgment", "title": "今日判断", "content": "秋粮能否顺利归仓，还要看后期病虫害防控；AI行程能否走得通，要看营业时间、道路耗时是否及时更新，也要区分资料过时与路线规划错误。" }, { "id": "macro", "title": "宏观／金融／政策", "items": [ { "id": "macro-grain", "title": "秋粮收获过三成，后期防灾仍影响最终收成", "summary": "10月3日《新闻联播》引述农业农村部农情调度称，全国秋粮收获已过三成，进度快于去年同期，集中收获期预计持续到10月下旬。报道同时提醒，长江流域近期气温偏高增加稻飞虱风险；收获进度领先还不能当作全年粮食增产的统计结果。", "category": "macro", "sources": [ { "name": "央视网《新闻联播》", "url": "https://tv.cctv.com/2026/10/03/VIDEkfYUUuA5wFaYGd6CpyMs261003.shtml" } ], "published_at": "2026-10-03T19:11:00+08:00", "source_name": "央视网《新闻联播》", "source_url": "https://tv.cctv.com/2026/10/03/VIDEkfYUUuA5wFaYGd6CpyMs261003.shtml" }, { "id": "macro-dumping", "title": "商务部对欧盟对硝基甲苯启动反倾销调查", "summary": "商务部10月3日宣布，应国内产业申请，对原产于欧盟的进口对硝基甲苯发起反倾销调查。发言人称，申请人的初步证据显示相关进口价格在2022至2025年累计下跌近60%；这仍是立案依据，尚非最终倾销裁决，也不等于已经加征反倾销税。", "category": "macro", "sources": [ { "name": "商务部新闻办公室", "url": "https://www.mofcom.gov.cn/xwfb/xwfyrth/art/2026/art_53e95cd6d6b64c43be04d11cbf37aa1c.html" } ], "published_at": "2026-10-03T10:15:00+08:00", "source_name": "商务部新闻办公室", "source_url": "https://www.mofcom.gov.cn/xwfb/xwfyrth/art/2026/art_53e95cd6d6b64c43be04d11cbf37aa1c.html" }, { "id": "g20-browser-body", "title": "G20贸易部长会议只就粮食议题形成联合声明", "summary": "商务部10月2日晚披露，9月30日至10月1日举行的G20贸易部长会议就粮食议题达成联合声明，其他议题未形成成果文件。中方在会上主张维护最惠国待遇原则；不能把会议讨论调整该原则，理解为各方已同意改写贸易规则。", "category": "macro", "sources": [ { "name": "商务部新闻办公室", "url": "https://www.mofcom.gov.cn/xwfb/bldhd/art/2026/art_4fe7620f820b4ef1a3a053981d79d2ef.html" } ], "published_at": "2026-10-02T22:10:00+08:00", "source_name": "商务部新闻办公室", "source_url": "https://www.mofcom.gov.cn/xwfb/bldhd/art/2026/art_4fe7620f820b4ef1a3a053981d79d2ef.html" } ] }, { "id": "hotspots", "title": "中文互联网热点／讨论场", "items": [ { "id": "ai-travel-oct3-interviews", "title": "AI旅游攻略的短板，出在行程执行时", "fact_description": "10月3日，定焦One刊发五名旅行者的采访，新浪财经同日转载。这组采访回顾了中秋、国庆及更早的旅行经历。受访者陈亭称，中秋赴长白山时，AI估计的50分钟车程实际用了1小时50分钟；玲玲称，韩国行程安排上午逛汉南洞，到场才发现大多店铺中午营业。两段经历分别涉及道路耗时与商铺营业时段：行程表上的时间看似衔接，到现场却出现了车程延长和扑空，打乱原定安排。", "discussion": "微博“凤凰周刊”把小众目的地拥挤归于推荐趋同，知乎“赵泠”则质疑训练资料与网上信息的时效。前者关注游客是否被引向同一地点，后者关注攻略采用的资料是否跟得上现场变化。", "boundary": "这组采访没有受控测试，无法估计各产品失败率，也不能证明推荐趋同造成拥堵；缺少完整提示词和工具调用记录，资料过时与路线计算错误仍难区分。", "next_variables": "核对同一路线的回答时间与实测耗时；若信息新鲜仍持续绕路，应转查规划能力。目的地拥挤还需比较推荐曝光与到访变化。", "near_model": { "name": "信息时效与状态估计", "field": "信息系统", "text": "行程规划依赖当前状态。地点坐标变化慢，营业情况和道路耗时却变化快；把历史资料当成即时状态，即使路线推理正确，也会给出不可执行的安排。以营业时间为例，若资料仍是旧时段，重新排列景点顺序也解决不了扑空；更新状态后，才有必要继续检查交通衔接。" }, "far_model": { "name": "生态学的理想自由分布", "field": "行为生态学", "text": "理想自由分布讨论个体怎样随资源收益选择去处：某处收益高便吸引更多个体，拥挤又降低收益。把清静程度视作旅行收益，广泛推荐会改变“小众”的条件。但游客缺少完整信息，也不能无成本改行程，现实不满足模型理想假设。它对近模型的修正是：即使资料刚刚更新，共同推荐本身也可能改变现场。" }, "category": "hotspot", "sources": [ { "name": "新浪财经转载定焦One", "url": "https://finance.sina.com.cn/roll/2026-10-03/doc-initwzsm0894631.shtml" }, { "name": "微博／凤凰周刊", "url": "https://weibo.com/1267454277/RkWk6cmFr" }, { "name": "知乎／赵泠", "url": "https://www.zhihu.com/question/2089851178997608583/answer/2089922787880212249" } ], "freshness": { "type": "new_reporting", "published_at": "2026-10-03T09:26:00+08:00", "increment": "五名旅行者新采访报道；经历跨中秋、国庆及更早旅行，不统称新事故" } } ] }, { "id": "local-life", "title": "沪杭今天的活动（10月4日）", "summary": "以下为沪杭精选活动，并非完整清单。票价与限制按售票页列示，不保证余票、即时出票或现场名额。售票截止不等于最晚入场；付款前请复核场次及入场安排。 表内按开始时间排序，同一时间保留编排顺序，不表示推荐高低。", "city_refund_note": "本表活动售出票品不退换；平台列明的例外为因不可抗力导致演出取消或延期。", "items": [ { "title": "怪兽计划乐队『隔在你我之间的那片海』2026巡演 上海站", "time": "20:00—21:30", "venue": "上海 MAO Livehouse", "address": "上海市黄浦区重庆南路308号3楼", "price": "预售票 70元", "restrictions": "售出不退换（不可抗力例外见表旁说明）；预售票售票至2026-10-04 19:00:00", "why": "", "id": "event-shanghai-1", "category": "event", "city": "上海", "sources": [ { "name": "秀动活动详情", "url": "https://www.showstart.com/event/307836" } ], "date": "2026-10-04", "source_name": "秀动活动详情", "source_url": "https://www.showstart.com/event/307836" }, { "title": "澳大利亚INFP系梦幻治愈组合DREAM TUNES巡演 上海站", "time": "20:00—21:20", "venue": "育音堂小镇 C厅 YUYINTOWN CUBE ", "address": "上海市长宁区昭化路658号CPARK.D栋 地下一层", "price": "预售票 198元", "restrictions": "售票页对艺人国别的标注不一致；180元早鸟及340元双人早鸟（两人整套）售票至今日10:00，余票以购票页为准；售出不退换（不可抗力例外见表旁说明）；预售票售票至2026-10-04 21:00:00", "why": "", "id": "event-shanghai-2", "category": "event", "city": "上海", "sources": [ { "name": "秀动活动详情", "url": "https://www.showstart.com/event/309692" } ], "date": "2026-10-04", "source_name": "秀动活动详情", "source_url": "https://www.showstart.com/event/309692" }, { "title": "张黎2026「欢迎光临」巡演上海站", "time": "20:00—22:00", "venue": "7HERTZ LIVE", "address": "上海市闵行区虹许路731号-4号楼", "price": "普通票 128元", "restrictions": "另列9.90元早鸟，余票不明；售出不退换（不可抗力例外见表旁说明）；普通票售票至2026-10-04 22:00:00", "why": "", "id": "event-shanghai-3", "category": "event", "city": "上海", "sources": [ { "name": "秀动活动详情", "url": "https://www.showstart.com/event/310084" } ], "date": "2026-10-04", "source_name": "秀动活动详情", "source_url": "https://www.showstart.com/event/310084" }, { "title": "LINKIN PARK 林肯公园&新金属 摇滚致敬音乐会", "time": "20:30—23:00", "venue": "珍珠剧场 THE PEARL", "address": "上海市虹口区乍浦路471号靠近武进路", "price": "预售票 150元", "restrictions": "18岁及以上，禁止未成年人入场。由剧场驻场外籍乐队演绎，非LINKIN PARK本人出演；凭平台电子票二维码入场，一人一码。票价仅含当晚演出，不保证座位；散台先到先得，可能拼桌；卡座／包厢另有低消；售出不退换（不可抗力例外见表旁说明）；预售票售票至2026-10-04 19:00:00；开门时间18:00", "why": "", "id": "event-shanghai-4", "category": "event", "city": "上海", "sources": [ { "name": "秀动活动详情", "url": "https://www.showstart.com/event/310018" } ], "date": "2026-10-04", "source_name": "秀动活动详情", "source_url": "https://www.showstart.com/event/310018" }, { "title": "「Back to the Other Side」巴西当代爵士HII TRIO 杭州站", "time": "19:30—21:00", "venue": "爵士黄楼", "address": "杭州市上城区柳营路6号", "price": "单人预售 120元", "restrictions": "19:00开始入场；儿童与成人同价；按到场先后选座；谢绝自带酒水、饮料、食物；双人预售200元为两人整套价格，售票至今日19:30；售出不退换（不可抗力例外见表旁说明）；单人预售售票至2026-10-04 19:30:00", "why": "演出期间保持安静，手机静音，摄影关闭闪光灯；曲目间或中场休息时离席。", "id": "event-hangzhou-1", "category": "event", "city": "杭州", "sources": [ { "name": "秀动活动详情", "url": "https://www.showstart.com/event/310376" } ], "date": "2026-10-04", "source_name": "秀动活动详情", "source_url": "https://www.showstart.com/event/310376" }, { "title": "《四月是你的谎言》钢琴小提琴唯美经典音乐会", "time": "19:30—21:00", "venue": "浙江省文化馆小剧场", "address": "杭州市拱墅区武林路77号小剧场", "price": "票价 100元", "restrictions": "售出不退换（不可抗力例外见表旁说明）；100元售票至2026-10-04 21:00:00", "why": "", "id": "event-hangzhou-2", "category": "event", "city": "杭州", "sources": [ { "name": "秀动活动详情", "url": "https://www.showstart.com/event/309201" } ], "date": "2026-10-04", "source_name": "秀动活动详情", "source_url": "https://www.showstart.com/event/309201" }, { "title": "帆布小镇今夜无法入眠Ⅱ特别专场-杭州站", "time": "20:00—21:40", "venue": " CH8-LIVEHOUSE（杭州小河店）", "address": "浙江省杭州市拱墅区拱宸桥街道小河路118号", "price": "全价票 150元", "restrictions": "99元早鸟与120元预售已于今日00:00结束；建议开演前约30分钟进场；售出不退换（不可抗力例外见表旁说明）；全价票售票至2026-10-04 22:00:00", "why": "现场有乐队周边售卖，演出结束后有签售环节。", "id": "event-hangzhou-3", "category": "event", "city": "杭州", "sources": [ { "name": "秀动活动详情", "url": "https://www.showstart.com/event/306730" } ], "date": "2026-10-04", "source_name": "秀动活动详情", "source_url": "https://www.showstart.com/event/306730" } ] }, { "id": "papers", "title": "今日新论文", "summary": "周末补读：arXiv当前仍为10月2日周五新刊，以下三篇此前未刊。新刊日期不等于提交日期；结果均为作者报告，尚非独立复现。", "items": [ { "id": "2610.01306", "title": "DAYJOB: A Benchmark for Long-Horizon Professional Work", "summary": "DAYJOB收集医疗与金融领域的130项专业任务，让代理从简短要求和混杂资料中完成交付。作者报告，所测最强配置Claude Opus 5.5在医疗与金融任务中的严格通过率分别为24.7%和23.9%；这里要求每项评分标准全部达标，并由代理裁判评阅，不是一般问答正确率。论文中的错误案例包括接受病历已经否定的前提，提醒使用者检查输入依据，而不只审阅最终文字。", "version": "v1", "category": "paper", "sources": [ { "name": "arXiv作者预印本", "url": "https://arxiv.org/abs/2610.01306v1" } ], "published_at": "2026-10-02（arXiv新刊列表日期，无精确时刻）", "editorial_title": "DAYJOB把专业任务的通过门槛设为全部要求达标", "listing_date": "2026-10-02", "authors": [ "Finley, Stephanie", "Panavas, Liudas", "Mikkelson, Thomas", "Hinton, Cam", "Ganss, Stacey", "Monton, Bradley", "Kendall, Emily", "Spradlin, Michelle", "Bye, Lydia", "O'Brien, Michael", "Ylvisaker, Lauren", "Ray, Derek", "Garre, Suhaas", "Mehta, Sushant", "Chen, Edwin" ], "submitted_time_utc": "2026-10-01T08:39:10+00:00", "submitted_time_cst": "2026-10-01T16:39:10+08:00", "listing_header": "Showing new listings for Friday, 2 October 2026", "listing_section": "New submissions", "freshness_basis": "listing_date_weekend_carryover", "tags": [ "专业任务", "评测" ], "source_name": "arXiv作者预印本", "source_url": "https://arxiv.org/abs/2610.01306v1" }, { "id": "2610.00651", "title": "Agent Evaluation Reliability: More Tasks Won't (Always) Fix An Agent Leaderboard", "summary": "这篇研究用贝叶斯方差分解检查代理排行榜，区分“模型加运行框架”的整体排名与模型本身的排名。作者发现，前者稳定，并不保证后者也稳定；当不确定性主要来自只测了少数运行框架时，继续增加同类任务仍留下排名差异。选型若要判断模型能力，应让同一模型跨框架测试。论文结论依赖当前样本，排名稳定也不等于评测内容有效。", "version": "v1", "category": "paper", "sources": [ { "name": "arXiv作者预印本", "url": "https://arxiv.org/abs/2610.00651v1" } ], "published_at": "2026-10-02（arXiv新刊列表日期，无精确时刻）", "editorial_title": "代理榜单的模型排名受运行框架影响", "listing_date": "2026-10-02", "authors": [ "Hardy, Michael", "Azam, Ruhana", "Reuel, Anka", "Kochenderfer, Mykel", "Koyejo, Sanmi" ], "submitted_time_utc": "2026-09-30T19:52:09+00:00", "submitted_time_cst": "2026-10-01T03:52:09+08:00", "listing_header": "Showing new listings for Friday, 2 October 2026", "listing_section": "New submissions", "freshness_basis": "listing_date_weekend_carryover", "tags": [ "代理评测", "运行框架" ], "source_name": "arXiv作者预印本", "source_url": "https://arxiv.org/abs/2610.00651v1" }, { "id": "2610.02070", "title": "Causal Memory Policy: Making Memory Utility Identifiable by Intervening on Retrieval", "summary": "Causal Memory Policy为部分上下文位置随机安排记忆，检验长期未被检索的内容到底有没有用。在LongMemEval独立子集上，作者报告区分必要与非必要记忆的AUC由0.542升至0.664。但实验候选池使用了必要记忆的已知标签，实际系统如何找到这个池仍未解决；对当前问题有用，也不足以决定未来该删除什么。", "version": "v1", "category": "paper", "sources": [ { "name": "arXiv作者预印本", "url": "https://arxiv.org/abs/2610.02070v1" } ], "published_at": "2026-10-02（arXiv新刊列表日期，无精确时刻）", "editorial_title": "检索不到的记忆，不能凭使用记录判成无用", "listing_date": "2026-10-02", "authors": [ "Behnam, Arman", "Wang, Binghui" ], "submitted_time_utc": "2026-10-01T17:11:25+00:00", "submitted_time_cst": "2026-10-02T01:11:25+08:00", "listing_header": "Showing new listings for Friday, 2 October 2026", "listing_section": "New submissions", "freshness_basis": "listing_date_weekend_carryover", "tags": [ "长期记忆", "检索" ], "source_name": "arXiv作者预印本", "source_url": "https://arxiv.org/abs/2610.02070v1" } ] }, { "id": "ai", "title": "AI／开发者工具", "items": [ { "id": "kolibri", "title": "Aleph Alpha开放Kolibri权重，主攻德英双语应用", "summary": "Aleph Alpha于10月3日开放Kolibri模型权重，采用Apache 2.0许可证，面向德语和英语任务。官方介绍其总参数约78B、每个token激活约3B，支持最长100万token上下文；有本地部署需求的团队多了一个可评测选项，但不能按3B稠密模型估算显存，也不能由厂商基准直接判断中文表现。", "category": "ai_product", "sources": [ { "name": "Aleph Alpha官方博客", "url": "https://aleph-alpha.com/en/blog/kolibri-has-landed-a-sovereign-open-weight-model/" } ], "published_at": "2026-10-03（页面日期，未列时区或时分）", "source_name": "Aleph Alpha官方博客", "source_url": "https://aleph-alpha.com/en/blog/kolibri-has-landed-a-sovereign-open-weight-model/" }, { "id": "claude-exact", "title": "Claude Code新版本修复文件读取与Shell权限检查漏洞", "summary": "Claude Code v2.1.289于北京时间10月4日07:07发布，修复了通过符号链接在IDE中引用文件时，Read拒绝规则未生效的问题；沙箱自动放行模式下，部分带变量赋值前缀的Shell命令也曾漏过拒绝或询问规则。依赖这些限制管理代理权限的团队，应核对升级后的规则行为，不能只看配置文件里是否写了拒绝项。", "category": "ai_product", "sources": [ { "name": "GitHub精确版本API", "url": "https://api.github.com/repos/anthropics/claude-code/releases/tags/v2.1.289" }, { "name": "Anthropic GitHub release页面", "url": "https://github.com/anthropics/claude-code/releases/tag/v2.1.289" } ], "published_at": "2026-10-04T07:07:17+08:00", "source_name": "GitHub精确版本API", "source_url": "https://api.github.com/repos/anthropics/claude-code/releases/tags/v2.1.289" } ] } ], "items": [ { "id": "macro-grain", "title": "秋粮收获过三成，后期防灾仍影响最终收成", "summary": "10月3日《新闻联播》引述农业农村部农情调度称，全国秋粮收获已过三成，进度快于去年同期，集中收获期预计持续到10月下旬。报道同时提醒，长江流域近期气温偏高增加稻飞虱风险；收获进度领先还不能当作全年粮食增产的统计结果。", "category": "macro", "sources": [ { "name": "央视网《新闻联播》", "url": "https://tv.cctv.com/2026/10/03/VIDEkfYUUuA5wFaYGd6CpyMs261003.shtml" } ], "published_at": "2026-10-03T19:11:00+08:00", "source_name": "央视网《新闻联播》", "source_url": "https://tv.cctv.com/2026/10/03/VIDEkfYUUuA5wFaYGd6CpyMs261003.shtml" }, { "id": "macro-dumping", "title": "商务部对欧盟对硝基甲苯启动反倾销调查", "summary": "商务部10月3日宣布，应国内产业申请，对原产于欧盟的进口对硝基甲苯发起反倾销调查。发言人称，申请人的初步证据显示相关进口价格在2022至2025年累计下跌近60%；这仍是立案依据，尚非最终倾销裁决，也不等于已经加征反倾销税。", "category": "macro", "sources": [ { "name": "商务部新闻办公室", "url": "https://www.mofcom.gov.cn/xwfb/xwfyrth/art/2026/art_53e95cd6d6b64c43be04d11cbf37aa1c.html" } ], "published_at": "2026-10-03T10:15:00+08:00", "source_name": "商务部新闻办公室", "source_url": "https://www.mofcom.gov.cn/xwfb/xwfyrth/art/2026/art_53e95cd6d6b64c43be04d11cbf37aa1c.html" }, { "id": "g20-browser-body", "title": "G20贸易部长会议只就粮食议题形成联合声明", "summary": "商务部10月2日晚披露，9月30日至10月1日举行的G20贸易部长会议就粮食议题达成联合声明，其他议题未形成成果文件。中方在会上主张维护最惠国待遇原则；不能把会议讨论调整该原则，理解为各方已同意改写贸易规则。", "category": "macro", "sources": [ { "name": "商务部新闻办公室", "url": "https://www.mofcom.gov.cn/xwfb/bldhd/art/2026/art_4fe7620f820b4ef1a3a053981d79d2ef.html" } ], "published_at": "2026-10-02T22:10:00+08:00", "source_name": "商务部新闻办公室", "source_url": "https://www.mofcom.gov.cn/xwfb/bldhd/art/2026/art_4fe7620f820b4ef1a3a053981d79d2ef.html" }, { "id": "ai-travel-oct3-interviews", "title": "AI旅游攻略的短板，出在行程执行时", "fact_description": "10月3日，定焦One刊发五名旅行者的采访，新浪财经同日转载。这组采访回顾了中秋、国庆及更早的旅行经历。受访者陈亭称，中秋赴长白山时，AI估计的50分钟车程实际用了1小时50分钟；玲玲称，韩国行程安排上午逛汉南洞，到场才发现大多店铺中午营业。两段经历分别涉及道路耗时与商铺营业时段：行程表上的时间看似衔接，到现场却出现了车程延长和扑空，打乱原定安排。", "discussion": "微博“凤凰周刊”把小众目的地拥挤归于推荐趋同，知乎“赵泠”则质疑训练资料与网上信息的时效。前者关注游客是否被引向同一地点，后者关注攻略采用的资料是否跟得上现场变化。", "boundary": "这组采访没有受控测试，无法估计各产品失败率，也不能证明推荐趋同造成拥堵；缺少完整提示词和工具调用记录，资料过时与路线计算错误仍难区分。", "next_variables": "核对同一路线的回答时间与实测耗时；若信息新鲜仍持续绕路，应转查规划能力。目的地拥挤还需比较推荐曝光与到访变化。", "near_model": { "name": "信息时效与状态估计", "field": "信息系统", "text": "行程规划依赖当前状态。地点坐标变化慢，营业情况和道路耗时却变化快；把历史资料当成即时状态，即使路线推理正确，也会给出不可执行的安排。以营业时间为例，若资料仍是旧时段，重新排列景点顺序也解决不了扑空；更新状态后，才有必要继续检查交通衔接。" }, "far_model": { "name": "生态学的理想自由分布", "field": "行为生态学", "text": "理想自由分布讨论个体怎样随资源收益选择去处：某处收益高便吸引更多个体，拥挤又降低收益。把清静程度视作旅行收益，广泛推荐会改变“小众”的条件。但游客缺少完整信息，也不能无成本改行程，现实不满足模型理想假设。它对近模型的修正是：即使资料刚刚更新，共同推荐本身也可能改变现场。" }, "category": "hotspot", "sources": [ { "name": "新浪财经转载定焦One", "url": "https://finance.sina.com.cn/roll/2026-10-03/doc-initwzsm0894631.shtml" }, { "name": "微博／凤凰周刊", "url": "https://weibo.com/1267454277/RkWk6cmFr" }, { "name": "知乎／赵泠", "url": "https://www.zhihu.com/question/2089851178997608583/answer/2089922787880212249" } ], "freshness": { "type": "new_reporting", "published_at": "2026-10-03T09:26:00+08:00", "increment": "五名旅行者新采访报道；经历跨中秋、国庆及更早旅行，不统称新事故" } }, { "title": "怪兽计划乐队『隔在你我之间的那片海』2026巡演 上海站", "time": "20:00—21:30", "venue": "上海 MAO Livehouse", "address": "上海市黄浦区重庆南路308号3楼", "price": "预售票 70元", "restrictions": "售出不退换（不可抗力例外见表旁说明）；预售票售票至2026-10-04 19:00:00", "why": "", "id": "event-shanghai-1", "category": "event", "city": "上海", "sources": [ { "name": "秀动活动详情", "url": "https://www.showstart.com/event/307836" } ], "date": "2026-10-04", "source_name": "秀动活动详情", "source_url": "https://www.showstart.com/event/307836" }, { "title": "澳大利亚INFP系梦幻治愈组合DREAM TUNES巡演 上海站", "time": "20:00—21:20", "venue": "育音堂小镇 C厅 YUYINTOWN CUBE ", "address": "上海市长宁区昭化路658号CPARK.D栋 地下一层", "price": "预售票 198元", "restrictions": "售票页对艺人国别的标注不一致；180元早鸟及340元双人早鸟（两人整套）售票至今日10:00，余票以购票页为准；售出不退换（不可抗力例外见表旁说明）；预售票售票至2026-10-04 21:00:00", "why": "", "id": "event-shanghai-2", "category": "event", "city": "上海", "sources": [ { "name": "秀动活动详情", "url": "https://www.showstart.com/event/309692" } ], "date": "2026-10-04", "source_name": "秀动活动详情", "source_url": "https://www.showstart.com/event/309692" }, { "title": "张黎2026「欢迎光临」巡演上海站", "time": "20:00—22:00", "venue": "7HERTZ LIVE", "address": "上海市闵行区虹许路731号-4号楼", "price": "普通票 128元", "restrictions": "另列9.90元早鸟，余票不明；售出不退换（不可抗力例外见表旁说明）；普通票售票至2026-10-04 22:00:00", "why": "", "id": "event-shanghai-3", "category": "event", "city": "上海", "sources": [ { "name": "秀动活动详情", "url": "https://www.showstart.com/event/310084" } ], "date": "2026-10-04", "source_name": "秀动活动详情", "source_url": "https://www.showstart.com/event/310084" }, { "title": "LINKIN PARK 林肯公园&新金属 摇滚致敬音乐会", "time": "20:30—23:00", "venue": "珍珠剧场 THE PEARL", "address": "上海市虹口区乍浦路471号靠近武进路", "price": "预售票 150元", "restrictions": "18岁及以上，禁止未成年人入场。由剧场驻场外籍乐队演绎，非LINKIN PARK本人出演；凭平台电子票二维码入场，一人一码。票价仅含当晚演出，不保证座位；散台先到先得，可能拼桌；卡座／包厢另有低消；售出不退换（不可抗力例外见表旁说明）；预售票售票至2026-10-04 19:00:00；开门时间18:00", "why": "", "id": "event-shanghai-4", "category": "event", "city": "上海", "sources": [ { "name": "秀动活动详情", "url": "https://www.showstart.com/event/310018" } ], "date": "2026-10-04", "source_name": "秀动活动详情", "source_url": "https://www.showstart.com/event/310018" }, { "title": "「Back to the Other Side」巴西当代爵士HII TRIO 杭州站", "time": "19:30—21:00", "venue": "爵士黄楼", "address": "杭州市上城区柳营路6号", "price": "单人预售 120元", "restrictions": "19:00开始入场；儿童与成人同价；按到场先后选座；谢绝自带酒水、饮料、食物；双人预售200元为两人整套价格，售票至今日19:30；售出不退换（不可抗力例外见表旁说明）；单人预售售票至2026-10-04 19:30:00", "why": "演出期间保持安静，手机静音，摄影关闭闪光灯；曲目间或中场休息时离席。", "id": "event-hangzhou-1", "category": "event", "city": "杭州", "sources": [ { "name": "秀动活动详情", "url": "https://www.showstart.com/event/310376" } ], "date": "2026-10-04", "source_name": "秀动活动详情", "source_url": "https://www.showstart.com/event/310376" }, { "title": "《四月是你的谎言》钢琴小提琴唯美经典音乐会", "time": "19:30—21:00", "venue": "浙江省文化馆小剧场", "address": "杭州市拱墅区武林路77号小剧场", "price": "票价 100元", "restrictions": "售出不退换（不可抗力例外见表旁说明）；100元售票至2026-10-04 21:00:00", "why": "", "id": "event-hangzhou-2", "category": "event", "city": "杭州", "sources": [ { "name": "秀动活动详情", "url": "https://www.showstart.com/event/309201" } ], "date": "2026-10-04", "source_name": "秀动活动详情", "source_url": "https://www.showstart.com/event/309201" }, { "title": "帆布小镇今夜无法入眠Ⅱ特别专场-杭州站", "time": "20:00—21:40", "venue": " CH8-LIVEHOUSE（杭州小河店）", "address": "浙江省杭州市拱墅区拱宸桥街道小河路118号", "price": "全价票 150元", "restrictions": "99元早鸟与120元预售已于今日00:00结束；建议开演前约30分钟进场；售出不退换（不可抗力例外见表旁说明）；全价票售票至2026-10-04 22:00:00", "why": "现场有乐队周边售卖，演出结束后有签售环节。", "id": "event-hangzhou-3", "category": "event", "city": "杭州", "sources": [ { "name": "秀动活动详情", "url": "https://www.showstart.com/event/306730" } ], "date": "2026-10-04", "source_name": "秀动活动详情", "source_url": "https://www.showstart.com/event/306730" }, { "id": "2610.01306", "title": "DAYJOB: A Benchmark for Long-Horizon Professional Work", "summary": "DAYJOB收集医疗与金融领域的130项专业任务，让代理从简短要求和混杂资料中完成交付。作者报告，所测最强配置Claude Opus 5.5在医疗与金融任务中的严格通过率分别为24.7%和23.9%；这里要求每项评分标准全部达标，并由代理裁判评阅，不是一般问答正确率。论文中的错误案例包括接受病历已经否定的前提，提醒使用者检查输入依据，而不只审阅最终文字。", "version": "v1", "category": "paper", "sources": [ { "name": "arXiv作者预印本", "url": "https://arxiv.org/abs/2610.01306v1" } ], "published_at": "2026-10-02（arXiv新刊列表日期，无精确时刻）", "editorial_title": "DAYJOB把专业任务的通过门槛设为全部要求达标", "listing_date": "2026-10-02", "authors": [ "Finley, Stephanie", "Panavas, Liudas", "Mikkelson, Thomas", "Hinton, Cam", "Ganss, Stacey", "Monton, Bradley", "Kendall, Emily", "Spradlin, Michelle", "Bye, Lydia", "O'Brien, Michael", "Ylvisaker, Lauren", "Ray, Derek", "Garre, Suhaas", "Mehta, Sushant", "Chen, Edwin" ], "submitted_time_utc": "2026-10-01T08:39:10+00:00", "submitted_time_cst": "2026-10-01T16:39:10+08:00", "listing_header": "Showing new listings for Friday, 2 October 2026", "listing_section": "New submissions", "freshness_basis": "listing_date_weekend_carryover", "tags": [ "专业任务", "评测" ], "source_name": "arXiv作者预印本", "source_url": "https://arxiv.org/abs/2610.01306v1" }, { "id": "2610.00651", "title": "Agent Evaluation Reliability: More Tasks Won't (Always) Fix An Agent Leaderboard", "summary": "这篇研究用贝叶斯方差分解检查代理排行榜，区分“模型加运行框架”的整体排名与模型本身的排名。作者发现，前者稳定，并不保证后者也稳定；当不确定性主要来自只测了少数运行框架时，继续增加同类任务仍留下排名差异。选型若要判断模型能力，应让同一模型跨框架测试。论文结论依赖当前样本，排名稳定也不等于评测内容有效。", "version": "v1", "category": "paper", "sources": [ { "name": "arXiv作者预印本", "url": "https://arxiv.org/abs/2610.00651v1" } ], "published_at": "2026-10-02（arXiv新刊列表日期，无精确时刻）", "editorial_title": "代理榜单的模型排名受运行框架影响", "listing_date": "2026-10-02", "authors": [ "Hardy, Michael", "Azam, Ruhana", "Reuel, Anka", "Kochenderfer, Mykel", "Koyejo, Sanmi" ], "submitted_time_utc": "2026-09-30T19:52:09+00:00", "submitted_time_cst": "2026-10-01T03:52:09+08:00", "listing_header": "Showing new listings for Friday, 2 October 2026", "listing_section": "New submissions", "freshness_basis": "listing_date_weekend_carryover", "tags": [ "代理评测", "运行框架" ], "source_name": "arXiv作者预印本", "source_url": "https://arxiv.org/abs/2610.00651v1" }, { "id": "2610.02070", "title": "Causal Memory Policy: Making Memory Utility Identifiable by Intervening on Retrieval", "summary": "Causal Memory Policy为部分上下文位置随机安排记忆，检验长期未被检索的内容到底有没有用。在LongMemEval独立子集上，作者报告区分必要与非必要记忆的AUC由0.542升至0.664。但实验候选池使用了必要记忆的已知标签，实际系统如何找到这个池仍未解决；对当前问题有用，也不足以决定未来该删除什么。", "version": "v1", "category": "paper", "sources": [ { "name": "arXiv作者预印本", "url": "https://arxiv.org/abs/2610.02070v1" } ], "published_at": "2026-10-02（arXiv新刊列表日期，无精确时刻）", "editorial_title": "检索不到的记忆，不能凭使用记录判成无用", "listing_date": "2026-10-02", "authors": [ "Behnam, Arman", "Wang, Binghui" ], "submitted_time_utc": "2026-10-01T17:11:25+00:00", "submitted_time_cst": "2026-10-02T01:11:25+08:00", "listing_header": "Showing new listings for Friday, 2 October 2026", "listing_section": "New submissions", "freshness_basis": "listing_date_weekend_carryover", "tags": [ "长期记忆", "检索" ], "source_name": "arXiv作者预印本", "source_url": "https://arxiv.org/abs/2610.02070v1" }, { "id": "kolibri", "title": "Aleph Alpha开放Kolibri权重，主攻德英双语应用", "summary": "Aleph Alpha于10月3日开放Kolibri模型权重，采用Apache 2.0许可证，面向德语和英语任务。官方介绍其总参数约78B、每个token激活约3B，支持最长100万token上下文；有本地部署需求的团队多了一个可评测选项，但不能按3B稠密模型估算显存，也不能由厂商基准直接判断中文表现。", "category": "ai_product", "sources": [ { "name": "Aleph Alpha官方博客", "url": "https://aleph-alpha.com/en/blog/kolibri-has-landed-a-sovereign-open-weight-model/" } ], "published_at": "2026-10-03（页面日期，未列时区或时分）", "source_name": "Aleph Alpha官方博客", "source_url": "https://aleph-alpha.com/en/blog/kolibri-has-landed-a-sovereign-open-weight-model/" }, { "id": "claude-exact", "title": "Claude Code新版本修复文件读取与Shell权限检查漏洞", "summary": "Claude Code v2.1.289于北京时间10月4日07:07发布，修复了通过符号链接在IDE中引用文件时，Read拒绝规则未生效的问题；沙箱自动放行模式下，部分带变量赋值前缀的Shell命令也曾漏过拒绝或询问规则。依赖这些限制管理代理权限的团队，应核对升级后的规则行为，不能只看配置文件里是否写了拒绝项。", "category": "ai_product", "sources": [ { "name": "GitHub精确版本API", "url": "https://api.github.com/repos/anthropics/claude-code/releases/tags/v2.1.289" }, { "name": "Anthropic GitHub release页面", "url": "https://github.com/anthropics/claude-code/releases/tag/v2.1.289" } ], "published_at": "2026-10-04T07:07:17+08:00", "source_name": "GitHub精确版本API", "source_url": "https://api.github.com/repos/anthropics/claude-code/releases/tags/v2.1.289" } ], "local_life": { "shanghai": [ { "title": "怪兽计划乐队『隔在你我之间的那片海』2026巡演 上海站", "time": "20:00—21:30", "venue": "上海 MAO Livehouse", "address": "上海市黄浦区重庆南路308号3楼", "price": "预售票 70元", "restrictions": "售出不退换（不可抗力例外见表旁说明）；预售票售票至2026-10-04 19:00:00", "why": "", "id": "event-shanghai-1", "category": "event", "city": "上海", "sources": [ { "name": "秀动活动详情", "url": "https://www.showstart.com/event/307836" } ], "date": "2026-10-04", "source_name": "秀动活动详情", "source_url": "https://www.showstart.com/event/307836" }, { "title": "澳大利亚INFP系梦幻治愈组合DREAM TUNES巡演 上海站", "time": "20:00—21:20", "venue": "育音堂小镇 C厅 YUYINTOWN CUBE ", "address": "上海市长宁区昭化路658号CPARK.D栋 地下一层", "price": "预售票 198元", "restrictions": "售票页对艺人国别的标注不一致；180元早鸟及340元双人早鸟（两人整套）售票至今日10:00，余票以购票页为准；售出不退换（不可抗力例外见表旁说明）；预售票售票至2026-10-04 21:00:00", "why": "", "id": "event-shanghai-2", "category": "event", "city": "上海", "sources": [ { "name": "秀动活动详情", "url": "https://www.showstart.com/event/309692" } ], "date": "2026-10-04", "source_name": "秀动活动详情", "source_url": "https://www.showstart.com/event/309692" }, { "title": "张黎2026「欢迎光临」巡演上海站", "time": "20:00—22:00", "venue": "7HERTZ LIVE", "address": "上海市闵行区虹许路731号-4号楼", "price": "普通票 128元", "restrictions": "另列9.90元早鸟，余票不明；售出不退换（不可抗力例外见表旁说明）；普通票售票至2026-10-04 22:00:00", "why": "", "id": "event-shanghai-3", "category": "event", "city": "上海", "sources": [ { "name": "秀动活动详情", "url": "https://www.showstart.com/event/310084" } ], "date": "2026-10-04", "source_name": "秀动活动详情", "source_url": "https://www.showstart.com/event/310084" }, { "title": "LINKIN PARK 林肯公园&新金属 摇滚致敬音乐会", "time": "20:30—23:00", "venue": "珍珠剧场 THE PEARL", "address": "上海市虹口区乍浦路471号靠近武进路", "price": "预售票 150元", "restrictions": "18岁及以上，禁止未成年人入场。由剧场驻场外籍乐队演绎，非LINKIN PARK本人出演；凭平台电子票二维码入场，一人一码。票价仅含当晚演出，不保证座位；散台先到先得，可能拼桌；卡座／包厢另有低消；售出不退换（不可抗力例外见表旁说明）；预售票售票至2026-10-04 19:00:00；开门时间18:00", "why": "", "id": "event-shanghai-4", "category": "event", "city": "上海", "sources": [ { "name": "秀动活动详情", "url": "https://www.showstart.com/event/310018" } ], "date": "2026-10-04", "source_name": "秀动活动详情", "source_url": "https://www.showstart.com/event/310018" } ], "hangzhou": [ { "title": "「Back to the Other Side」巴西当代爵士HII TRIO 杭州站", "time": "19:30—21:00", "venue": "爵士黄楼", "address": "杭州市上城区柳营路6号", "price": "单人预售 120元", "restrictions": "19:00开始入场；儿童与成人同价；按到场先后选座；谢绝自带酒水、饮料、食物；双人预售200元为两人整套价格，售票至今日19:30；售出不退换（不可抗力例外见表旁说明）；单人预售售票至2026-10-04 19:30:00", "why": "演出期间保持安静，手机静音，摄影关闭闪光灯；曲目间或中场休息时离席。", "id": "event-hangzhou-1", "category": "event", "city": "杭州", "sources": [ { "name": "秀动活动详情", "url": "https://www.showstart.com/event/310376" } ], "date": "2026-10-04", "source_name": "秀动活动详情", "source_url": "https://www.showstart.com/event/310376" }, { "title": "《四月是你的谎言》钢琴小提琴唯美经典音乐会", "time": "19:30—21:00", "venue": "浙江省文化馆小剧场", "address": "杭州市拱墅区武林路77号小剧场", "price": "票价 100元", "restrictions": "售出不退换（不可抗力例外见表旁说明）；100元售票至2026-10-04 21:00:00", "why": "", "id": "event-hangzhou-2", "category": "event", "city": "杭州", "sources": [ { "name": "秀动活动详情", "url": "https://www.showstart.com/event/309201" } ], "date": "2026-10-04", "source_name": "秀动活动详情", "source_url": "https://www.showstart.com/event/309201" }, { "title": "帆布小镇今夜无法入眠Ⅱ特别专场-杭州站", "time": "20:00—21:40", "venue": " CH8-LIVEHOUSE（杭州小河店）", "address": "浙江省杭州市拱墅区拱宸桥街道小河路118号", "price": "全价票 150元", "restrictions": "99元早鸟与120元预售已于今日00:00结束；建议开演前约30分钟进场；售出不退换（不可抗力例外见表旁说明）；全价票售票至2026-10-04 22:00:00", "why": "现场有乐队周边售卖，演出结束后有签售环节。", "id": "event-hangzhou-3", "category": "event", "city": "杭州", "sources": [ { "name": "秀动活动详情", "url": "https://www.showstart.com/event/306730" } ], "date": "2026-10-04", "source_name": "秀动活动详情", "source_url": "https://www.showstart.com/event/306730" } ] }, "papers": [ { "id": "2610.01306", "title": "DAYJOB: A Benchmark for Long-Horizon Professional Work", "summary": "DAYJOB收集医疗与金融领域的130项专业任务，让代理从简短要求和混杂资料中完成交付。作者报告，所测最强配置Claude Opus 5.5在医疗与金融任务中的严格通过率分别为24.7%和23.9%；这里要求每项评分标准全部达标，并由代理裁判评阅，不是一般问答正确率。论文中的错误案例包括接受病历已经否定的前提，提醒使用者检查输入依据，而不只审阅最终文字。", "version": "v1", "category": "paper", "sources": [ { "name": "arXiv作者预印本", "url": "https://arxiv.org/abs/2610.01306v1" } ], "published_at": "2026-10-02（arXiv新刊列表日期，无精确时刻）", "editorial_title": "DAYJOB把专业任务的通过门槛设为全部要求达标", "listing_date": "2026-10-02", "authors": [ "Finley, Stephanie", "Panavas, Liudas", "Mikkelson, Thomas", "Hinton, Cam", "Ganss, Stacey", "Monton, Bradley", "Kendall, Emily", "Spradlin, Michelle", "Bye, Lydia", "O'Brien, Michael", "Ylvisaker, Lauren", "Ray, Derek", "Garre, Suhaas", "Mehta, Sushant", "Chen, Edwin" ], "submitted_time_utc": "2026-10-01T08:39:10+00:00", "submitted_time_cst": "2026-10-01T16:39:10+08:00", "listing_header": "Showing new listings for Friday, 2 October 2026", "listing_section": "New submissions", "freshness_basis": "listing_date_weekend_carryover", "tags": [ "专业任务", "评测" ], "source_name": "arXiv作者预印本", "source_url": "https://arxiv.org/abs/2610.01306v1" }, { "id": "2610.00651", "title": "Agent Evaluation Reliability: More Tasks Won't (Always) Fix An Agent Leaderboard", "summary": "这篇研究用贝叶斯方差分解检查代理排行榜，区分“模型加运行框架”的整体排名与模型本身的排名。作者发现，前者稳定，并不保证后者也稳定；当不确定性主要来自只测了少数运行框架时，继续增加同类任务仍留下排名差异。选型若要判断模型能力，应让同一模型跨框架测试。论文结论依赖当前样本，排名稳定也不等于评测内容有效。", "version": "v1", "category": "paper", "sources": [ { "name": "arXiv作者预印本", "url": "https://arxiv.org/abs/2610.00651v1" } ], "published_at": "2026-10-02（arXiv新刊列表日期，无精确时刻）", "editorial_title": "代理榜单的模型排名受运行框架影响", "listing_date": "2026-10-02", "authors": [ "Hardy, Michael", "Azam, Ruhana", "Reuel, Anka", "Kochenderfer, Mykel", "Koyejo, Sanmi" ], "submitted_time_utc": "2026-09-30T19:52:09+00:00", "submitted_time_cst": "2026-10-01T03:52:09+08:00", "listing_header": "Showing new listings for Friday, 2 October 2026", "listing_section": "New submissions", "freshness_basis": "listing_date_weekend_carryover", "tags": [ "代理评测", "运行框架" ], "source_name": "arXiv作者预印本", "source_url": "https://arxiv.org/abs/2610.00651v1" }, { "id": "2610.02070", "title": "Causal Memory Policy: Making Memory Utility Identifiable by Intervening on Retrieval", "summary": "Causal Memory Policy为部分上下文位置随机安排记忆，检验长期未被检索的内容到底有没有用。在LongMemEval独立子集上，作者报告区分必要与非必要记忆的AUC由0.542升至0.664。但实验候选池使用了必要记忆的已知标签，实际系统如何找到这个池仍未解决；对当前问题有用，也不足以决定未来该删除什么。", "version": "v1", "category": "paper", "sources": [ { "name": "arXiv作者预印本", "url": "https://arxiv.org/abs/2610.02070v1" } ], "published_at": "2026-10-02（arXiv新刊列表日期，无精确时刻）", "editorial_title": "检索不到的记忆，不能凭使用记录判成无用", "listing_date": "2026-10-02", "authors": [ "Behnam, Arman", "Wang, Binghui" ], "submitted_time_utc": "2026-10-01T17:11:25+00:00", "submitted_time_cst": "2026-10-02T01:11:25+08:00", "listing_header": "Showing new listings for Friday, 2 October 2026", "listing_section": "New submissions", "freshness_basis": "listing_date_weekend_carryover", "tags": [ "长期记忆", "检索" ], "source_name": "arXiv作者预印本", "source_url": "https://arxiv.org/abs/2610.02070v1" } ], "sources": [ { "name": "央视网《新闻联播》", "url": "https://tv.cctv.com/2026/10/03/VIDEkfYUUuA5wFaYGd6CpyMs261003.shtml" }, { "name": "商务部新闻办公室", "url": "https://www.mofcom.gov.cn/xwfb/xwfyrth/art/2026/art_53e95cd6d6b64c43be04d11cbf37aa1c.html" }, { "name": "商务部新闻办公室", "url": "https://www.mofcom.gov.cn/xwfb/bldhd/art/2026/art_4fe7620f820b4ef1a3a053981d79d2ef.html" }, { "name": "新浪财经转载定焦One", "url": "https://finance.sina.com.cn/roll/2026-10-03/doc-initwzsm0894631.shtml" }, { "name": "微博／凤凰周刊", "url": "https://weibo.com/1267454277/RkWk6cmFr" }, { "name": "知乎／赵泠", "url": "https://www.zhihu.com/question/2089851178997608583/answer/2089922787880212249" }, { "name": "秀动活动详情", "url": "https://www.showstart.com/event/307836" }, { "name": "秀动活动详情", "url": "https://www.showstart.com/event/309692" }, { "name": "秀动活动详情", "url": "https://www.showstart.com/event/310084" }, { "name": "秀动活动详情", "url": "https://www.showstart.com/event/310018" }, { "name": "秀动活动详情", "url": "https://www.showstart.com/event/310376" }, { "name": "秀动活动详情", "url": "https://www.showstart.com/event/309201" }, { "name": "秀动活动详情", "url": "https://www.showstart.com/event/306730" }, { "name": "arXiv作者预印本", "url": "https://arxiv.org/abs/2610.01306v1" }, { "name": "arXiv作者预印本", "url": "https://arxiv.org/abs/2610.00651v1" }, { "name": "arXiv作者预印本", "url": "https://arxiv.org/abs/2610.02070v1" }, { "name": "Aleph Alpha官方博客", "url": "https://aleph-alpha.com/en/blog/kolibri-has-landed-a-sovereign-open-weight-model/" }, { "name": "GitHub精确版本API", "url": "https://api.github.com/repos/anthropics/claude-code/releases/tags/v2.1.289" }, { "name": "Anthropic GitHub release页面", "url": "https://github.com/anthropics/claude-code/releases/tag/v2.1.289" } ], "local_life_preview": "上海精选4场；杭州精选3场。出发前请查看活动表中的时间、票价与入场限制。" }