[{"data":1,"prerenderedAt":822},["ShallowReactive",2],{"page-/post/nezus/2026/07/ai":3,"pages-grouped-only":343,"surrounding-page":817},{"id":4,"title":5,"author":6,"body":7,"date":332,"description":18,"extension":333,"group":334,"lastmod":335,"meta":336,"navigation":337,"path":338,"rawbody":339,"seo":340,"showTitle":335,"sort":335,"stem":341,"tags":335,"versions":335,"__hash__":342},"content/post/nezus/2026/07/ai.md","你刷到的帖子可能全是AI写的","storytelling",{"type":8,"value":9,"toc":318},"minimark",[10,15,19,22,25,28,31,34,37,45,52,59,62,65,71,78,81,84,87,90,94,97,104,110,113,116,119,126,131,134,138,144,155,160,163,169,173,176,179,186,196,202,205,210,214,220,223,226,229,232,235,241,247,250,255,258,262,265,268,271,277,280,283,286,289,292,295,300,303,309,312],[11,12,14],"h2",{"id":13},"你刷到的帖子可能有一半是-ai-写的","你刷到的帖子，可能有一半是 AI 写的",[16,17,18],"p",{},"周五晚上十一点，你躺在床上刷手机。",[16,20,21],{},"打开 LinkedIn ，一条行业分析跳进视线——\"2026 年数字化转型的五大趋势\"。开头讲了个小故事，引用了麦肯锡的数据，结尾还抛出一个「发人深省」的问题。排版干净，措辞得体，像是一个资深从业者花了两个小时打磨出来的。",[16,23,24],{},"你点了个赞，划走了。",[16,26,27],{},"但你可能不知道：那篇帖子，作者连键盘都没碰。",[11,29,30],{"id":30},"一份让全网沉默的数据",[16,32,33],{},"安全公司 Pangram 做了一个实验。",[16,35,36],{},"他们开发了一款 Chrome 浏览器插件，用户在刷社交媒体的时候，插件会在后台悄悄扫描每一条帖子，用 AI 检测模型判断它是不是机器写的。用户可以选择匿名分享扫描数据，用于研究。",[16,38,39,40,44],{},"两个月后，他们拿到了一份惊人的数据集：",[41,42,43],"strong",{},"超过 100 万条帖子","，覆盖 LinkedIn、X/Twitter、Reddit、Medium、Substack 等主流平台。",[16,46,47,48,51],{},"结论是什么？四个字：",[41,49,50],{},"触目惊心","。",[16,53,54,55,58],{},"整体来看，所有被扫描的帖子中，",[41,56,57],{},"13.8% 被标记为 AI 生成","。相当于你每刷七八条内容，就有一条根本不是人写的。",[16,60,61],{},"但这只是平均值。如果你多看几眼那些「看起来很有营养」的长文章，比例直接翻倍。",[11,63,64],{"id":64},"长文才是重灾区",[16,66,67,68,51],{},"Pangram 的分析发现了一个规律：",[41,69,70],{},"内容越长，越可能是 AI 写的",[16,72,73,74,77],{},"在所有超过 250 个单词的长帖子中，",[41,75,76],{},"25.72% 完全由 AI 生成","。四分之一。你看到的每一篇行业分析、经验分享、深度思考，每四篇里就有一篇是机器吐出来的。",[16,79,80],{},"想想你上周收藏的那些「干货」文章。",[16,82,83],{},"其中至少四分之一的作者，可能只做了一件事：把标题扔进 ChatGPT ，复制粘贴。",[16,85,86],{},"唯一的例外是 Substack。长文和短文的 AI 比例差不多持平——这大概是因为 Substack 上的作者要靠付费订阅吃饭， AI 糊弄不了愿意掏钱的读者。",[16,88,89],{},"但其他平台呢？情况比你想的糟得多。",[11,91,93],{"id":92},"linkedin-ai-写作的法外之地","LinkedIn ： AI 写作的「法外之地」",[16,95,96],{},"LinkedIn 是这场 AI 洪水中的「决堤口」。",[16,98,99,100,103],{},"数据显示， LinkedIn 上超过 40% 的长文帖子被标记为",[41,101,102],{},"完全 AI 生成","。不是「部分润色」，不是「AI 辅助」——是整个人都不在场的那种「完全」。",[16,105,106,107,51],{},"更夸张的是： LinkedIn 的帖子数量只占全部扫描数据的三分之一，但它贡献了",[41,108,109],{},"所有 AI 标记内容中的 62%",[16,111,112],{},"三分之二的 AI 垃圾，都堆在一个本该最「专业」的地方。",[16,114,115],{},"讽刺吗？一个以「真实身份」「职业信誉」为核心卖点的平台，反而成了 AI 最泛滥的角落。人们宁愿用一个假的声音，去维护一个真的名字。",[16,117,118],{},"LinkedIn 自己也没少助力。平台上内置的「AI 润色」按钮——从「用 AI 写」改名叫「优化帖子」，换汤不换药——无时无刻不在鼓励用户：别费劲了，让 AI 替你说话。",[16,120,121,122,125],{},"更黑色幽默的是， LinkedIn 的一位高管最近发帖宣布平台将",[41,123,124],{},"检测并降权 AI 生成的帖子","。猜猜那条宣布打击 AI 的帖子本身是什么检测结果？",[16,127,128],{},[41,129,130],{},"AI 生成的。",[16,132,133],{},"自己打自己脸，打得啪啪响。",[11,135,137],{"id":136},"xtwitter-一半内容不是人写的","X/Twitter ：一半内容不是人写的",[16,139,140,141,51],{},"如果说 LinkedIn 的问题是「长文全是 AI」，那 X/Twitter 的问题更彻底：",[41,142,143],{},"全平台沦陷",[16,145,146,147,150,151,154],{},"Pangram 的数据显示， X 上的文章中，",[41,148,149],{},"23.9% 完全由 AI 生成","，另外 ",[41,152,153],{},"22.9% 是 AI 和人类混合写作","。两项加起来——46.8%。",[16,156,157],{},[41,158,159],{},"只有 53.2% 的 X 文章被标记为「完全人类撰写」。",[16,161,162],{},"你刷 Twitter 看新闻、看观点、看吵架，每两条里就有一条跟 AI 有关。你以为自己在了解「大家怎么看」，实际上你看到的「大家」，可能是一堆 token。",[16,164,165,166,51],{},"2024 年有人开玩笑说「互联网已经死了」。2026 年的数据告诉你：",[41,167,168],{},"它可能真的快死了",[11,170,172],{"id":171},"reddit-最后一片人类净土","Reddit ：最后一片「人类净土」？",[16,174,175],{},"在所有平台中， Reddit 的数据最好看——整体 AI 率只有 4.4%。",[16,177,178],{},"但别高兴太早。",[16,180,181,182,185],{},"Reddit 的「干净」有一个取巧的原因：",[41,183,184],{},"回复太多","。Reddit 被扫描的内容中， 72% 是评论回复，而这些回复 98.1% 是人类写的。大量短回复「lol」「this」「based」把分母撑大了，拉低了整体比例。",[16,187,188,189,192,193,51],{},"如果你只看",[41,190,191],{},"顶层帖子","——那些被推到首页、获得上千点赞的内容——AI 率立刻跳到 ",[41,194,195],{},"11.6%",[16,197,198,199,51],{},"更诡异的是，即使控制字数变量， Reddit 的顶层帖子仍然是 AI 生成的概率比评论高出 ",[41,200,201],{},"5.25 倍",[16,203,204],{},"这说明什么？ Reddit 的反机器人策略——限制发帖频率、要求人工验证——确实管用了。它成功地让 AI 无法用「自动回复」的方式灌水。但它拦不住那些用 AI 写一篇「真诚分享」再手动发出去的人。",[16,206,207],{},[41,208,209],{},"低强度的 AI 作弊，完美地绕过了所有防御。",[11,211,213],{"id":212},"一个-001-的细节","一个 0.01% 的细节",[16,215,216,217,51],{},"在继续往下聊之前，有一个数字你必须知道：",[41,218,219],{},"0.01%",[16,221,222],{},"这是 Pangram 3.3 模型的假阳性率——把人类写的文章错误地标记为 AI 生成的几率，只有万分之一。",[16,224,225],{},"换句话说，上面那些数据不是「可能误判」，是真的有那么多 AI 内容。当你觉得「AI 检测不准吧」的时候，这个模型告诉你：准到你可以信。",[11,227,228],{"id":228},"这件事为什么细思极恐",[16,230,231],{},"我们不妨做一道简单的算术。",[16,233,234],{},"假设你每天花 30 分钟刷社交媒体，平均看 100 条帖子。按照 13.8% 的整体 AI 率，其中 14 条是 AI 写的。如果你偏爱看长文章，这个数字可能翻倍到 28 条。",[16,236,237,238,51],{},"一个月下来，你「消费」了 ",[41,239,240],{},"400 多条 AI 生产的信息垃圾",[16,242,243,244],{},"一年呢？",[41,245,246],{},"5000 条。",[16,248,249],{},"而你甚至不知道它们是谁写的、为什么写、背后的动机是什么。它们可能是营销号在批量生产流量内容，可能是某个公司在做「品牌建设」，也可能是某个组织在进行舆论引导。",[16,251,252],{},[41,253,254],{},"你以为是信息获取，其实是被投喂。",[16,256,257],{},"更可怕的是 AI 内容的「隐蔽进化」。早期的 ChatGPT 文章一眼就能看出来——「在当今数字化时代」「总而言之」「值得注意的是」。但现在的模型，尤其是专门针对社交媒体优化的写作工具，产出的文字已经越来越像真人。Pangram 的研究人员甚至发现，很多 AI 内容会被刻意加入一些「小瑕疵」——故意拼错一个词、用口语化表达——来逃过检测。",[11,259,261],{"id":260},"我们该怎么办","我们该怎么办？",[16,263,264],{},"说实话，没有完美的解决方案。",[16,266,267],{},"平台在行动——LinkedIn 说要降权 AI 帖子， Reddit 在加强验证机制。但这些措施要么是「自己也在用 AI 写公告」的尴尬表演，要么只拦得住最低级的灌水行为。",[16,269,270],{},"AI 检测工具像 Pangram 的插件一样，至少给了你一个「知情权」。知道一条内容可能是 AI 写的，和不知道，是两种完全不同的阅读体验。前者叫「浏览」，后者叫「被喂养」。",[16,272,273,274],{},"但更根本的问题在于：",[41,275,276],{},"我们为什么刷社交媒体？",[16,278,279],{},"如果是为了获取真实的人类经验和观点，那 AI 内容就是毒药。如果只是无聊打发时间，那谁写的可能确实无所谓——反正都是划过。",[16,281,282],{},"最危险的是中间地带：你以为你在了解世界，实际上你看到的世界，是一群语言模型拼凑出来的幻象。",[11,284,285],{"id":285},"下次刷手机的时候",[16,287,288],{},"当你再打开 LinkedIn ，看到那篇「2026 年你必须知道的五个趋势」。",[16,290,291],{},"当你再打开 Twitter ，看到那篇「深度分析：为什么 XX 注定失败」。",[16,293,294],{},"当你再打开 Reddit ，看到那篇「作为一个从业十年的老兵，我想说……」",[16,296,297],{},[41,298,299],{},"停一秒。",[16,301,302],{},"问问自己：这是一个人花时间写出来的真实经验，还是一个模型在 3 秒内吐出来的概率组合？",[16,304,305,306],{},"你不需要每次都猜对。你只需要知道：",[41,307,308],{},"你刷到的帖子，可能有一半根本不是人写的。",[16,310,311],{},"知道这件事本身，就已经比 99% 的人更清醒了。",[16,313,314,317],{},[41,315,316],{},"封面高亮关键词",": AI 写的, 一半",{"title":319,"searchDepth":320,"depth":320,"links":321},"",2,[322,323,324,325,326,327,328,329,330,331],{"id":13,"depth":320,"text":14},{"id":30,"depth":320,"text":30},{"id":64,"depth":320,"text":64},{"id":92,"depth":320,"text":93},{"id":136,"depth":320,"text":137},{"id":171,"depth":320,"text":172},{"id":212,"depth":320,"text":213},{"id":228,"depth":320,"text":228},{"id":260,"depth":320,"text":261},{"id":285,"depth":320,"text":285},"2026-07-10T00:00:00.000Z","md","#202607AI资讯",null,{},true,"/post/nezus/2026/07/ai","---\ntitle: \"你刷到的帖子可能全是AI写的\"\ndate: 2026-07-10\ngroup: \"#202607AI资讯\"\ndescription: \"周五晚上十一点，你躺在床上刷手机。\"\nauthor: storytelling\n---\n\n\n## 你刷到的帖子，可能有一半是 AI 写的\n\n周五晚上十一点，你躺在床上刷手机。\n\n打开 LinkedIn ，一条行业分析跳进视线——\"2026 年数字化转型的五大趋势\"。开头讲了个小故事，引用了麦肯锡的数据，结尾还抛出一个「发人深省」的问题。排版干净，措辞得体，像是一个资深从业者花了两个小时打磨出来的。\n\n你点了个赞，划走了。\n\n但你可能不知道：那篇帖子，作者连键盘都没碰。\n\n## 一份让全网沉默的数据\n\n安全公司 Pangram 做了一个实验。\n\n他们开发了一款 Chrome 浏览器插件，用户在刷社交媒体的时候，插件会在后台悄悄扫描每一条帖子，用 AI 检测模型判断它是不是机器写的。用户可以选择匿名分享扫描数据，用于研究。\n\n两个月后，他们拿到了一份惊人的数据集：**超过 100 万条帖子**，覆盖 LinkedIn、X/Twitter、Reddit、Medium、Substack 等主流平台。\n\n结论是什么？四个字：**触目惊心**。\n\n整体来看，所有被扫描的帖子中，**13.8% 被标记为 AI 生成**。相当于你每刷七八条内容，就有一条根本不是人写的。\n\n但这只是平均值。如果你多看几眼那些「看起来很有营养」的长文章，比例直接翻倍。\n\n## 长文才是重灾区\n\nPangram 的分析发现了一个规律：**内容越长，越可能是 AI 写的**。\n\n在所有超过 250 个单词的长帖子中，**25.72% 完全由 AI 生成**。四分之一。你看到的每一篇行业分析、经验分享、深度思考，每四篇里就有一篇是机器吐出来的。\n\n想想你上周收藏的那些「干货」文章。\n\n其中至少四分之一的作者，可能只做了一件事：把标题扔进 ChatGPT ，复制粘贴。\n\n唯一的例外是 Substack。长文和短文的 AI 比例差不多持平——这大概是因为 Substack 上的作者要靠付费订阅吃饭， AI 糊弄不了愿意掏钱的读者。\n\n但其他平台呢？情况比你想的糟得多。\n\n## LinkedIn ： AI 写作的「法外之地」\n\nLinkedIn 是这场 AI 洪水中的「决堤口」。\n\n数据显示， LinkedIn 上超过 40% 的长文帖子被标记为**完全 AI 生成**。不是「部分润色」，不是「AI 辅助」——是整个人都不在场的那种「完全」。\n\n更夸张的是： LinkedIn 的帖子数量只占全部扫描数据的三分之一，但它贡献了**所有 AI 标记内容中的 62%**。\n\n三分之二的 AI 垃圾，都堆在一个本该最「专业」的地方。\n\n讽刺吗？一个以「真实身份」「职业信誉」为核心卖点的平台，反而成了 AI 最泛滥的角落。人们宁愿用一个假的声音，去维护一个真的名字。\n\nLinkedIn 自己也没少助力。平台上内置的「AI 润色」按钮——从「用 AI 写」改名叫「优化帖子」，换汤不换药——无时无刻不在鼓励用户：别费劲了，让 AI 替你说话。\n\n更黑色幽默的是， LinkedIn 的一位高管最近发帖宣布平台将**检测并降权 AI 生成的帖子**。猜猜那条宣布打击 AI 的帖子本身是什么检测结果？\n\n**AI 生成的。**\n\n自己打自己脸，打得啪啪响。\n\n## X/Twitter ：一半内容不是人写的\n\n如果说 LinkedIn 的问题是「长文全是 AI」，那 X/Twitter 的问题更彻底：**全平台沦陷**。\n\nPangram 的数据显示， X 上的文章中，**23.9% 完全由 AI 生成**，另外 **22.9% 是 AI 和人类混合写作**。两项加起来——46.8%。\n\n**只有 53.2% 的 X 文章被标记为「完全人类撰写」。**\n\n你刷 Twitter 看新闻、看观点、看吵架，每两条里就有一条跟 AI 有关。你以为自己在了解「大家怎么看」，实际上你看到的「大家」，可能是一堆 token。\n\n2024 年有人开玩笑说「互联网已经死了」。2026 年的数据告诉你：**它可能真的快死了**。\n\n## Reddit ：最后一片「人类净土」？\n\n在所有平台中， Reddit 的数据最好看——整体 AI 率只有 4.4%。\n\n但别高兴太早。\n\nReddit 的「干净」有一个取巧的原因：**回复太多**。Reddit 被扫描的内容中， 72% 是评论回复，而这些回复 98.1% 是人类写的。大量短回复「lol」「this」「based」把分母撑大了，拉低了整体比例。\n\n如果你只看**顶层帖子**——那些被推到首页、获得上千点赞的内容——AI 率立刻跳到 **11.6%**。\n\n更诡异的是，即使控制字数变量， Reddit 的顶层帖子仍然是 AI 生成的概率比评论高出 **5.25 倍**。\n\n这说明什么？ Reddit 的反机器人策略——限制发帖频率、要求人工验证——确实管用了。它成功地让 AI 无法用「自动回复」的方式灌水。但它拦不住那些用 AI 写一篇「真诚分享」再手动发出去的人。\n\n**低强度的 AI 作弊，完美地绕过了所有防御。**\n\n## 一个 0.01% 的细节\n\n在继续往下聊之前，有一个数字你必须知道：**0.01%**。\n\n这是 Pangram 3.3 模型的假阳性率——把人类写的文章错误地标记为 AI 生成的几率，只有万分之一。\n\n换句话说，上面那些数据不是「可能误判」，是真的有那么多 AI 内容。当你觉得「AI 检测不准吧」的时候，这个模型告诉你：准到你可以信。\n\n## 这件事为什么细思极恐\n\n我们不妨做一道简单的算术。\n\n假设你每天花 30 分钟刷社交媒体，平均看 100 条帖子。按照 13.8% 的整体 AI 率，其中 14 条是 AI 写的。如果你偏爱看长文章，这个数字可能翻倍到 28 条。\n\n一个月下来，你「消费」了 **400 多条 AI 生产的信息垃圾**。\n\n一年呢？**5000 条。**\n\n而你甚至不知道它们是谁写的、为什么写、背后的动机是什么。它们可能是营销号在批量生产流量内容，可能是某个公司在做「品牌建设」，也可能是某个组织在进行舆论引导。\n\n**你以为是信息获取，其实是被投喂。**\n\n更可怕的是 AI 内容的「隐蔽进化」。早期的 ChatGPT 文章一眼就能看出来——「在当今数字化时代」「总而言之」「值得注意的是」。但现在的模型，尤其是专门针对社交媒体优化的写作工具，产出的文字已经越来越像真人。Pangram 的研究人员甚至发现，很多 AI 内容会被刻意加入一些「小瑕疵」——故意拼错一个词、用口语化表达——来逃过检测。\n\n## 我们该怎么办？\n\n说实话，没有完美的解决方案。\n\n平台在行动——LinkedIn 说要降权 AI 帖子， Reddit 在加强验证机制。但这些措施要么是「自己也在用 AI 写公告」的尴尬表演，要么只拦得住最低级的灌水行为。\n\nAI 检测工具像 Pangram 的插件一样，至少给了你一个「知情权」。知道一条内容可能是 AI 写的，和不知道，是两种完全不同的阅读体验。前者叫「浏览」，后者叫「被喂养」。\n\n但更根本的问题在于：**我们为什么刷社交媒体？**\n\n如果是为了获取真实的人类经验和观点，那 AI 内容就是毒药。如果只是无聊打发时间，那谁写的可能确实无所谓——反正都是划过。\n\n最危险的是中间地带：你以为你在了解世界，实际上你看到的世界，是一群语言模型拼凑出来的幻象。\n\n## 下次刷手机的时候\n\n当你再打开 LinkedIn ，看到那篇「2026 年你必须知道的五个趋势」。\n\n当你再打开 Twitter ，看到那篇「深度分析：为什么 XX 注定失败」。\n\n当你再打开 Reddit ，看到那篇「作为一个从业十年的老兵，我想说……」\n\n**停一秒。**\n\n问问自己：这是一个人花时间写出来的真实经验，还是一个模型在 3 秒内吐出来的概率组合？\n\n你不需要每次都猜对。你只需要知道：**你刷到的帖子，可能有一半根本不是人写的。**\n\n知道这件事本身，就已经比 99% 的人更清醒了。\n\n**封面高亮关键词**: AI 写的, 一半",{"title":5,"description":18},"post/nezus/2026/07/ai","_nLZHjWc6uWAK2KUvrPkUIekBTGyvkxb-9yi8I0W-6k",[344,349,353,357,361,366,369,374,378,382,386,391,395,399,403,407,411,416,420,421,425,429,433,438,442,446,455,460,464,468,473,477,481,485,489,494,500,505,509,513,517,522,526,530,535,539,543,547,551,556,560,565,569,573,577,582,586,590,595,599,603,607,611,616,620,624,629,633,637,642,646,650,654,658,662,667,671,675,679,684,688,692,696,701,705,709,713,717,722,726,730,734,738,743,747,754,760,765,770,775,780,785,790,795,800,805,811],{"id":345,"path":346,"title":347,"date":348,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/2026-7-16.md","/post/nezus/2026/07/2026-7-16","2026年7月16日下午，台积电法说会上传出一个数字，让全球半导体圈炸了锅。","2026-07-17T00:00:00.000Z",{"id":350,"path":351,"title":352,"date":348,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/anthropic-ai.md","/post/nezus/2026/07/anthropic-ai","一个周末干掉四年：Anthropic 用 AI 两周搬完百万行代码，工程师沉默了",{"id":354,"path":355,"title":356,"date":348,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/google-20.md","/post/nezus/2026/07/google-20","欧盟一纸裁决，Google 20年护城河被撕开第一道口子",{"id":358,"path":359,"title":360,"date":348,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/xai-grok.md","/post/nezus/2026/07/xai-grok","xAI给Grok装上定时任务能力",{"id":362,"path":363,"title":364,"date":365,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/apple.md","/post/nezus/2026/07/apple","Apple智能落地中国，阿里千问成幕后赢家","2026-07-16T00:00:00.000Z",{"id":367,"path":368,"title":364,"date":365,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/deepmind.md","/post/nezus/2026/07/deepmind",{"id":370,"path":371,"title":372,"date":365,"tags":335,"group":334,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/07/gpt-vibe-coding-paste.md","/post/nezus/2026/07/gpt-vibe-coding-paste","用GPT Vibe Coding一个常用付费软件：Paste","user",{"id":375,"path":376,"title":377,"date":365,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/grok-build-agent.md","/post/nezus/2026/07/grok-build-agent","马斯克开源Grok Build，编程Agent进入白嫖时代",{"id":379,"path":380,"title":381,"date":365,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/openai-ai.md","/post/nezus/2026/07/openai-ai","OpenAI造了个AI黑客，专攻自家模型",{"id":383,"path":384,"title":385,"date":365,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/wps-comate-copilot.md","/post/nezus/2026/07/wps-comate-copilot","金山办公掏出WPS Comate，正面硬刚微软Copilot",{"id":387,"path":388,"title":389,"date":390,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/295b.md","/post/nezus/2026/07/295b","腾讯把295B大模型塞进了单张显卡","2026-07-15T00:00:00.000Z",{"id":392,"path":393,"title":394,"date":390,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/git-cursor.md","/post/nezus/2026/07/git-cursor","打开一个Git仓库Cursor就自动执行了恶意代码",{"id":396,"path":397,"title":398,"date":390,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/gpt-5-6-mac.md","/post/nezus/2026/07/gpt-5-6-mac","GPT-5.6偷偷删光了我的Mac文件",{"id":400,"path":401,"title":402,"date":390,"tags":335,"group":334,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/07/grok-cli-xai.md","/post/nezus/2026/07/grok-cli-xai","Grok CLI 偷传代码：xAI 把你的仓库打包上传了",{"id":404,"path":405,"title":402,"date":406,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/grok-cli-bug-done-xai.md","/post/nezus/2026/07/grok-cli-bug-done-xai","2026-07-14T00:00:00.000Z",{"id":408,"path":409,"title":410,"date":406,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/seedream-5-0-ai.md","/post/nezus/2026/07/seedream-5-0-ai","字节跳动Seedream 5.0：圈一下图片就改了，AI修图进入点选时代",{"id":412,"path":413,"title":414,"date":415,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/post-20260712-ddc9a8.md","/post/nezus/2026/07/post-20260712-ddc9a8","欧盟强推Chat Control 2.0：加密通信的生死时刻","2026-07-12T00:00:00.000Z",{"id":417,"path":418,"title":419,"date":332,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/ai-1-5.md","/post/nezus/2026/07/ai-1-5","AI烧了1.5万亿，能赚回来吗",{"id":4,"path":338,"title":5,"date":332,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},{"id":422,"path":423,"title":424,"date":332,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/chatgpt-openai.md","/post/nezus/2026/07/chatgpt-openai","ChatGPT学会自己干活了",{"id":426,"path":427,"title":428,"date":332,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/musk-anthropic-apology.md","/post/nezus/2026/07/musk-anthropic-apology","马斯克向Anthropic低头认错",{"id":430,"path":431,"title":432,"date":332,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/optimus.md","/post/nezus/2026/07/optimus","马斯克：Optimus不达标就开除团队",{"id":434,"path":435,"title":436,"date":437,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/anthropic-github.md","/post/nezus/2026/07/anthropic-github","Anthropic 花了一年打造的\"设计师灵魂\"，被一个 GitHub 仓库扒光了","2026-07-08T00:00:00.000Z",{"id":439,"path":440,"title":441,"date":437,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/meta-muse-image-video.md","/post/nezus/2026/07/meta-muse-image-video","扎克伯格掏出\"超级智能实验室\"第一张牌：30亿用户的社交数据，终于杀进了AI生图战场",{"id":443,"path":444,"title":445,"date":437,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/ms-mai-copilot.md","/post/nezus/2026/07/ms-mai-copilot","微软悄悄把OpenAI踢出局：Copilot正在用自研模型\"省钱\"，但你花的钱一分不少",{"id":447,"path":448,"title":449,"date":437,"tags":450,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/yc-ceo-37k-ai-code-bloat.md","/post/nezus/2026/07/yc-ceo-37k-ai-code-bloat","每天部署37000行代码，YC CEO的\"AI神迹\"被一个波兰程序员扒光了",[451,452,453,454],"AI","代码质量","YC","Garry Tan",{"id":456,"path":457,"title":458,"date":459,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/ai-1-6-gpt.md","/post/nezus/2026/07/ai-1-6-gpt","美团扔出一颗AI核弹：1.6万亿参数，代码能力碾压GPT","2026-07-06T00:00:00.000Z",{"id":461,"path":462,"title":463,"date":459,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/post-20260706-d12be3.md","/post/nezus/2026/07/post-20260706-d12be3","黄仁勋最重磅的新品，跳票了",{"id":465,"path":466,"title":467,"date":459,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/x-ai.md","/post/nezus/2026/07/x-ai","X 悄悄对 AI 敞开了大门，然后呢？",{"id":469,"path":470,"title":471,"date":472,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/30-2700.md","/post/nezus/2026/07/30-2700","AI Agent自主完成全球首例勒索攻击","2026-07-04T00:00:00.000Z",{"id":474,"path":475,"title":476,"date":472,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/ai-agent.md","/post/nezus/2026/07/ai-agent","万亿砸下去，AI Agent 还是没跑起来",{"id":478,"path":479,"title":480,"date":472,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/ai-ai.md","/post/nezus/2026/07/ai-ai","AI的身份证来了：网信办首次为AI服务设立专章",{"id":482,"path":483,"title":484,"date":472,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/anthropic-steganography-china.md","/post/nezus/2026/07/anthropic-steganography-china","Anthropic用隐写术识别中国用户：口口声声说AI安全的公司，背地里给你的代码下毒",{"id":486,"path":487,"title":488,"date":472,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/nvidia-rubin-ultra-cancelled.md","/post/nezus/2026/07/nvidia-rubin-ultra-cancelled","NVIDIA Rubin Ultra取消性能腰斩",{"id":490,"path":491,"title":492,"date":493,"tags":335,"group":334,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/07/keyboard.md","/post/nezus/2026/07/keyboard","Acti把AI Agent塞进手机键盘","2026-07-02T00:00:00.000Z",{"id":495,"path":496,"title":497,"date":498,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/ai-6-claude.md","/post/nezus/2026/06/ai-6-claude","四大AI决战文明6：Claude核平法国，还是输了","2026-06-29T00:00:00.000Z","#202606AI资讯",{"id":501,"path":502,"title":503,"date":504,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/deepseek.md","/post/nezus/2026/06/deepseek","DeepSeek成美国企业救命稻草，AI账单失控后100%切换","2026-06-28T00:00:00.000Z",{"id":506,"path":507,"title":508,"date":504,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/openai-gpt-5-6-20.md","/post/nezus/2026/06/openai-gpt-5-6-20","OpenAI发了最强模型，却被政府按住了：GPT-5.6只给20家公司用",{"id":510,"path":511,"title":512,"date":504,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/spacex-spacexai-xai-spacex.md","/post/nezus/2026/06/spacex-spacexai-xai-spacex","SpaceX 注册 SpaceXAI，马斯克要把 xAI 并入 SpaceX",{"id":514,"path":515,"title":516,"date":504,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/vision-openai-macbook.md","/post/nezus/2026/06/vision-openai-macbook","苹果Vision负责人跳槽OpenAI，触控MacBook要来了",{"id":518,"path":519,"title":520,"date":521,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/2026.md","/post/nezus/2026/06/2026","何小鹏放话：2026年底自动驾驶合法进全球","2026-06-27T00:00:00.000Z",{"id":523,"path":524,"title":525,"date":521,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/30-leaf-ai.md","/post/nezus/2026/06/30-leaf-ai","30秒克隆网红声音：Leaf开源AI分身项目",{"id":527,"path":528,"title":529,"date":521,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/400-ai.md","/post/nezus/2026/06/400-ai","400家报纸集体起诉：AI版权大战正式开打",{"id":531,"path":532,"title":533,"date":534,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/23-ai.md","/post/nezus/2026/06/23-ai","贝佐斯押注23亿美元，让AI用打游戏的方式学会控制机器人","2026-06-26T00:00:00.000Z",{"id":536,"path":537,"title":538,"date":534,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/2880-claude-ai.md","/post/nezus/2026/06/2880-claude-ai","2880万次\"偷师\"被当场抓包：Claude母公司一封信，撕开了中美AI暗战的遮羞布",{"id":540,"path":541,"title":542,"date":534,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/ai-1750-10-2.md","/post/nezus/2026/06/ai-1750-10-2","AI年收入1750亿美元，新增10亿收入只需2天",{"id":544,"path":545,"title":546,"date":534,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/ai.md","/post/nezus/2026/06/ai","深夜三点的红头文件：白宫为什么亲手掐死了全球最强AI？",{"id":548,"path":549,"title":550,"date":534,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/post-20260626-7a4a1d.md","/post/nezus/2026/06/post-20260626-7a4a1d","开源圈炸了！一个没人听过的模型，凭什么把编程天花板掀了？",{"id":552,"path":553,"title":554,"date":555,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/ai-google.md","/post/nezus/2026/06/ai-google","你的电脑，要被AI接管了：Google悄悄放了个大招","2026-06-25T00:00:00.000Z",{"id":557,"path":558,"title":559,"date":555,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/notion-cursor-ai.md","/post/nezus/2026/06/notion-cursor-ai","Notion把代码写交给了Cursor：一个文件夹就是一个AI员工",{"id":561,"path":562,"title":563,"date":564,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/alphafold-deepmind.md","/post/nezus/2026/06/alphafold-deepmind","AlphaFold之父离开DeepMind，背后藏着一盘更大的棋","2026-06-20T00:00:00.000Z",{"id":566,"path":567,"title":568,"date":564,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/cloudflare-ai-agent.md","/post/nezus/2026/06/cloudflare-ai-agent","Cloudflare给AI Agent发了身份证",{"id":570,"path":571,"title":572,"date":564,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/figure.md","/post/nezus/2026/06/figure","Figure工厂里，机器人比人多了",{"id":574,"path":575,"title":576,"date":564,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/pinecone-70.md","/post/nezus/2026/06/pinecone-70","阿里悄悄开源了内部数据库，Pinecone每月70美元要凉了？",{"id":578,"path":579,"title":580,"date":581,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/ai-50.md","/post/nezus/2026/06/ai-50","美国参议员要对AI公司征收50%的税了","2026-06-19T00:00:00.000Z",{"id":583,"path":584,"title":585,"date":581,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/claude-code.md","/post/nezus/2026/06/claude-code","Claude Code放大招：写完代码直接给你一个网页",{"id":587,"path":588,"title":589,"date":581,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/transformer-openai-27.md","/post/nezus/2026/06/transformer-openai-27","Transformer之父跳槽OpenAI：谷歌27亿美元打了水漂",{"id":591,"path":592,"title":593,"date":594,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/agent-vercel-eve-next-js-ai.md","/post/nezus/2026/06/agent-vercel-eve-next-js-ai","一个目录就是一个 Agent：Vercel 开源 Eve，Next.js 的故事要在 AI 重演？","2026-06-18T00:00:00.000Z",{"id":596,"path":597,"title":598,"date":594,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/g7.md","/post/nezus/2026/06/g7","硅谷教父在G7闭门会上说了一句话：把中国踢出去",{"id":600,"path":601,"title":602,"date":594,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/grok-4-3-amazon-bedrock.md","/post/nezus/2026/06/grok-4-3-amazon-bedrock","幻觉率全场最低，马斯克在AWS亮出了让对手沉默的底牌",{"id":604,"path":605,"title":606,"date":594,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/openai-930-2800.md","/post/nezus/2026/06/openai-930-2800","OpenAI年入930亿，却亏掉2800亿：这场豪赌，还能撑多久？",{"id":608,"path":609,"title":610,"date":594,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/vercel-ai.md","/post/nezus/2026/06/vercel-ai","Vercel悄悄扔了个炸弹：一个文件夹，就是一个AI员工",{"id":612,"path":613,"title":614,"date":615,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/5-160-tok-s-0-7-tok-s.md","/post/nezus/2026/06/5-160-tok-s-0-7-tok-s","5倍加速背后：160 tok/s起飞，0.7 tok/s坠毁","2026-06-17T00:00:00.000Z",{"id":617,"path":618,"title":619,"date":615,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/5.md","/post/nezus/2026/06/5","一个国产大模型，凌晨5点上线，然后把自己彻底交了出去",{"id":621,"path":622,"title":623,"date":615,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/deepseek-500.md","/post/nezus/2026/06/deepseek-500","DeepSeek首轮融资500亿：投资人无投票权",{"id":625,"path":626,"title":627,"date":628,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/15-ai.md","/post/nezus/2026/06/15-ai","15万人丢了饭碗，AI公司却在开香槟","2026-06-16T00:00:00.000Z",{"id":630,"path":631,"title":632,"date":628,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/kimi-k27-code-speed.md","/post/nezus/2026/06/kimi-k27-code-speed","速度快了6倍：Kimi K2.7 Code高速版，把编程AI卷进新赛道",{"id":634,"path":635,"title":636,"date":628,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/salesforce-36-ai-fin.md","/post/nezus/2026/06/salesforce-36-ai-fin","Salesforce 36亿美元买下AI客服公司Fin：巨头为何要为\"聊天机器人\"砸这么多钱？",{"id":638,"path":639,"title":640,"date":641,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/20.md","/post/nezus/2026/06/20","Meta被迫撤回20亿美元Manus收购，北京出手了","2026-06-15T00:00:00.000Z",{"id":643,"path":644,"title":645,"date":641,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/ai-openai-anthropic24.md","/post/nezus/2026/06/ai-openai-anthropic24","白宫AI监管被指偏袒OpenAI亚马逊，Anthropic24小时内被施压",{"id":647,"path":648,"title":649,"date":641,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/anthropic-dario-10--25.md","/post/nezus/2026/06/anthropic-dario-10-25","Anthropic上市前Dario自曝：文明崩溃概率10%-25%",{"id":651,"path":652,"title":653,"date":641,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/openai-1-5-ai.md","/post/nezus/2026/06/openai-1-5-ai","OpenAI砸1.5亿美元建合作伙伴网络，要搞AI界的安卓",{"id":655,"path":656,"title":657,"date":641,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/satya-nadella-ai.md","/post/nezus/2026/06/satya-nadella-ai","Satya Nadella：AI时代企业需同时构建人力资本和Token资本",{"id":659,"path":660,"title":661,"date":641,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/skill-ai.md","/post/nezus/2026/06/skill-ai","乔木小说创作Skill开源：AI自动生成剧情梗概人物设定",{"id":663,"path":664,"title":665,"date":666,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/anthropic.md","/post/nezus/2026/06/anthropic","Anthropic的困局：一边秘密上市，一边被政府召回","2026-06-14T00:00:00.000Z",{"id":668,"path":669,"title":670,"date":666,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/post-20260614-7ba4ce.md","/post/nezus/2026/06/post-20260614-7ba4ce","扎克伯格承认Meta AI转型脱轨：裁员10%、7000人转岗",{"id":672,"path":673,"title":674,"date":666,"tags":335,"group":499,"lastmod":335,"author":6,"sort":335},"content/post/nezus/2026/06/post-20260614-8e3b48.md","/post/nezus/2026/06/post-20260614-8e3b48","GLM-5.2全量开放：国产最强Coding模型来了",{"id":676,"path":677,"title":678,"date":666,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/post-20260614-ea1c30.md","/post/nezus/2026/06/post-20260614-ea1c30","谷歌安全负责人辞职：AI被用于军事，他拒绝沉默",{"id":680,"path":681,"title":682,"date":683,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/428b-minimax.md","/post/nezus/2026/06/428b-minimax","428B参数全部开源，MiniMax这步棋走对了","2026-06-13T00:00:00.000Z",{"id":685,"path":686,"title":687,"date":683,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/5-claude-anthropic-it.md","/post/nezus/2026/06/5-claude-anthropic-it","5万人用Claude办公，Anthropic拿下印度IT巨头",{"id":689,"path":690,"title":691,"date":683,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/agent.md","/post/nezus/2026/06/agent","豆包悄悄上了任务模式，字节的Agent野心藏不住了",{"id":693,"path":694,"title":695,"date":683,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/kimi-token-30.md","/post/nezus/2026/06/kimi-token-30","Kimi代码模型开源，推理token直接砍30%",{"id":697,"path":698,"title":699,"date":700,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/7-0-410.md","/post/nezus/2026/06/7-0-410","成立7个月，0产品，估值410亿美元——贝佐斯的新公司疯了","2026-06-12T00:00:00.000Z",{"id":702,"path":703,"title":704,"date":700,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/ai-ai.md","/post/nezus/2026/06/ai-ai","当造AI的人亲口说AI会抢走工作",{"id":706,"path":707,"title":708,"date":700,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/cursor-ai-ai.md","/post/nezus/2026/06/cursor-ai-ai","Cursor上新功能：AI动手前先让另一个AI批准",{"id":710,"path":711,"title":712,"date":700,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/midjourney.md","/post/nezus/2026/06/midjourney","你打开Midjourney的时候，它已经不是昨天的它了",{"id":714,"path":715,"title":716,"date":700,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/post-20260612-5071b7.md","/post/nezus/2026/06/post-20260612-5071b7","全自主无人机第一次杀了人，战争进入新纪元",{"id":718,"path":719,"title":720,"date":721,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/100-600.md","/post/nezus/2026/06/100-600","豆包说退票不到 100 块，结果花了 600——然后它帮用户写好了起诉自己的状子","2026-06-11T00:00:00.000Z",{"id":723,"path":724,"title":725,"date":721,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/180-ai-agent.md","/post/nezus/2026/06/180-ai-agent","180亿美元对冲基金炒掉所有分析师，改用数百个AI Agent炒股",{"id":727,"path":728,"title":729,"date":721,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/google-4.md","/post/nezus/2026/06/google-4","Google 开源 4 倍速扩散模型：自回归的黄昏？",{"id":731,"path":732,"title":733,"date":721,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/google-ai.md","/post/nezus/2026/06/google-ai","Google 因 AI '说错话'被判赔：幻觉的时代结束了？",{"id":735,"path":736,"title":737,"date":721,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/ip-ai.md","/post/nezus/2026/06/ip-ai","周星驰三部经典电影 IP 开放 AI 二创：火山方舟想做什么？",{"id":739,"path":740,"title":741,"date":742,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/gui.md","/post/nezus/2026/06/gui","人类需要GUI！","2026-06-09T00:00:00.000Z",{"id":744,"path":745,"title":746,"date":742,"tags":335,"group":499,"lastmod":335,"author":373,"sort":335},"content/post/nezus/2026/06/hermes-desktop.md","/post/nezus/2026/06/hermes-desktop","Hermes Desktop 安装及界面功能速览",{"id":748,"path":749,"title":750,"date":751,"tags":335,"group":752,"lastmod":753,"author":335,"sort":335},"content/post/ai/explore/01-interview-prompt-template.md","/post/ai/explore/01-interview-prompt-template","高级前端开发面试提示词模板","2026-01-21T07:30:52.902Z","面试SKILL","2026-01-21T07:31:28.343Z",{"id":755,"path":756,"title":757,"date":758,"tags":335,"group":759,"lastmod":758,"author":335,"sort":335},"content/post/ai/explore/03-interview-standard-answers.md","/post/ai/explore/03-interview-standard-answers","高级前端开发面试题标准答案","2026-01-21T07:25:00.000Z","面试SKILL:答案",{"id":761,"path":762,"title":763,"date":764,"tags":335,"group":759,"lastmod":764,"author":335,"sort":335},"content/post/ai/explore/03-10-team-management.md","/post/ai/explore/03-10-team-management","10.团队管理与技术债务 - 标准答案","2026-01-21T07:24:00.000Z",{"id":766,"path":767,"title":768,"date":769,"tags":335,"group":759,"lastmod":769,"author":335,"sort":335},"content/post/ai/explore/03-09-micro-frontend.md","/post/ai/explore/03-09-micro-frontend","9.微前端架构设计 - 标准答案","2026-01-21T07:23:00.000Z",{"id":771,"path":772,"title":773,"date":774,"tags":335,"group":759,"lastmod":774,"author":335,"sort":335},"content/post/ai/explore/03-08-nodejs-performance.md","/post/ai/explore/03-08-nodejs-performance","8.Node.js 性能调优 - 标准答案","2026-01-21T07:22:00.000Z",{"id":776,"path":777,"title":778,"date":779,"tags":335,"group":759,"lastmod":779,"author":335,"sort":335},"content/post/ai/explore/03-07-event-loop.md","/post/ai/explore/03-07-event-loop","7.Node.js Event Loop - 标准答案","2026-01-21T07:21:00.000Z",{"id":781,"path":782,"title":783,"date":784,"tags":335,"group":759,"lastmod":784,"author":335,"sort":335},"content/post/ai/explore/03-06-nuxt-ssr.md","/post/ai/explore/03-06-nuxt-ssr","6.Nuxt.js SSR 实战 - 标准答案","2026-01-21T07:20:00.000Z",{"id":786,"path":787,"title":788,"date":789,"tags":335,"group":759,"lastmod":789,"author":335,"sort":335},"content/post/ai/explore/03-05-vue-diff.md","/post/ai/explore/03-05-vue-diff","5.Vue Diff 算法 - 标准答案","2026-01-21T07:19:00.000Z",{"id":791,"path":792,"title":793,"date":794,"tags":335,"group":759,"lastmod":794,"author":335,"sort":335},"content/post/ai/explore/03-04-vue-migration.md","/post/ai/explore/03-04-vue-migration","4.Vue2 迁移到 Vue3 - 标准答案","2026-01-21T07:18:00.000Z",{"id":796,"path":797,"title":798,"date":799,"tags":335,"group":759,"lastmod":799,"author":335,"sort":335},"content/post/ai/explore/03-03-vue3-performance.md","/post/ai/explore/03-03-vue3-performance","3.Vue3 性能优化 - 标准答案","2026-01-21T07:17:00.000Z",{"id":801,"path":802,"title":803,"date":804,"tags":335,"group":759,"lastmod":804,"author":335,"sort":335},"content/post/ai/explore/03-02-component-communication.md","/post/ai/explore/03-02-component-communication","2.Vue 组件通信 - 标准答案","2026-01-21T07:16:00.000Z",{"id":806,"path":807,"title":808,"date":809,"tags":335,"group":759,"lastmod":810,"author":335,"sort":335},"content/post/ai/explore/03-01-vue-reactive.md","/post/ai/explore/03-01-vue-reactive","1.Vue 响应式原理 - 标准答案","2026-01-21T06:53:18.971Z","2026-01-21T07:15:00.000Z",{"id":812,"path":813,"title":814,"date":815,"tags":335,"group":752,"lastmod":816,"author":335,"sort":335},"content/post/ai/explore/02-interview-questions.md","/post/ai/explore/02-interview-questions","高级前端开发面试题目清单","2026-01-21T06:39:25.673Z","2026-01-21T07:12:45.186Z",[335,818],{"title":819,"path":820,"stem":821},"大模型，快就是好","/post/nezus/2026/08/post-20260807-58a2a9","post/nezus/2026/08/post-20260807-58a2a9",1786091424201]