ceepee retweeted
AI 裁员赌局开始反噬 $META
Meta 因员工大量修补 agent 的技术与安全错误，缩减裁员改用 AI 的计划并重新招回管理者。
Microsoft、McKinsey 限制员工用 AI，Walmart 员工称 AI 排班制造假警报。
Atlanta Fed 工作论文显示 90% 高管称 AI 尚未提升生产率，而今年美国已有逾 11.2 万裁员归因于 AI。
The AI layoff gamble is backfiring.
Companies that rushed to replace workers with AI are now cleaning up the mess, per The Guardian.
Meta scaled back plans to cut staff and use more AI agents after employees spent growing time cleaning up the agents’ technical and security errors. It’s now rehiring managers.
Microsoft and McKinsey are limiting employee AI use as costs pile up.
Walmart workers say AI scheduling and task tools set impossible standards and generate false alerts.
The payoff isn’t showing up: 90% of executives say AI hasn’t yet boosted productivity at their companies, per an Atlanta Fed working paper.
Yet more than 112,000 US job cuts this year have cited AI, per Challenger, Gray & Christmas.
ceepee retweeted
朋友带我去夜总会见世面。
他刚进包厢就搂着相熟的妹子唱歌，我坐在角落，看另一个姑娘给我倒了杯酒。
我说：“你不用陪我，我就是跟着来的。”
她点点头，把酒杯放到我面前：“好，那我坐这儿歇会儿。”
她问我做什么工作，我问她几点下班，聊到后来，她还出去帮我朋友挪了一次车。
我觉得她人挺好的，至少没劝我喝酒。
结账时，领班过来问朋友要不要带姑娘们出去吃点宵夜。
朋友搂着身边的妹子看着我没说话，我有点懵：“我不饿……”
姑娘正在穿外套，听见这话，转过身对我说：“那就谢谢老板了。”
我更愣神了，不知道她什么意思，这时领班递过账单：“老板大气，二位共消费1988块。”
到这一刻我才明白，在这种地方，收费的未必是你做了什么，也可能是你什么都没做。
ceepee retweeted
Google 宣布 K-12 学校场景下 Gemini、Gemini Notebook、Search AI 默认关闭，由学校逐个开启;
Search 的 AI Mode/AI Overviews 改为分步引导解题而非直接给答案，此举回应教育界对 AI 代写作业的反弹。
We want to clarify what we announced today about our education services. Today’s announcements give schools more choice, not less. Here’s what we announced:
- Gemini, Gemini Notebook, and Search will be OFF 🔴 by default. This is based on K-12 schools asking for more control over student use of AI services.
- Schools are in control: They can choose to turn ON 🟢 one, two, or all of these services individually.
- We are also updating Search for schools to make it better for learning. Specifically, AI Mode and AI Overviews will offer step-by-step guided learning through problems, so that students don’t get quick answers to schoolwork.
We know there’s concern about students using AI to shortcut schoolwork, and schools have told us they want control. That’s exactly why we are making these changes today to address the feedback we’ve heard from educators in recent months. And we’ll continue to engage deeply with educators to ensure we are giving schools more choice, more controls, and improved Search for learning.
Readers added context they thought people might want to know
Google's post claims schools can turn on Search independently from AI services like Gemini for more choice. Google executives stated Search and AI are deeply integrated, so schools enabling Search must accept AI features with no separate opt-out.
blog.google/products-and-p…
nytimes.com/2026/11/03/us/…
theverge.com/tech/1025411/g…
arstechnica.com/ai/2026/11/goo…
techcrunch.com/2026/11/03/goo…
androidauthority.com/google-to-turn…
ceepee retweeted
我做一了个 Agent 市场页面：放上头像、能力标签和一串数量，马上就显得很热闹。
有买家来问我：我雇它做这件事，真的比自己做更值吗？
BNB Chain 的 Build the Era 第一阶段有 260 多支团队参与。
进入后续 Set and Earn 阶段的 9 个市场里，就有 @termix_ai。
我觉得 TermiX 独立赛道的评判办法很有意思：
参赛者至少拿出 3 个真实任务，让 Agent 做一遍，自己再做一遍；把耗时、成本、交付质量和实际产出一起交上去。
TermiX 也会到参赛市场实际雇用 Agent，看最终交回什么。
这才是 Agent 市场该接受的检验。
一个 Agent 写着“会自动调仓”，买家只担心：任务有没有完成，花了多少钱，结果能不能用。
Set and Earn 将于 11 月 5 日 20:00（北京时间） 截止。
接下来这些入选市场能交出多少份令人满意的结果呢。
ceepee retweeted
最近看 Agent 项目，我都特别留意一个问题：Agent 做完自己手里任务之后，能不能主动找到下一个订单？
TermiX 今天演示了 Agent 在安装 TermiX Skills 后，通过 Binance Agent ID 互发消息。
这个入口让 Agent 之间有了直接联系的方式。
但接下来消息能否接上需求、服务匹配和市场交易：Agent 能否找到合适的服务方，把一项工作顺着流程推进下去。
TermiX 同时在做 Agent 间通信和可交易服务市场。
两条产品线都已能看到，接下来要验证的是它们之间能不能形成稳定的使用路径。
对 Agent Economy 来说，协作频次、从沟通到任务的转化，以及持续的市场收入，都会是重要指标。
DefiLlama 当前显示，TermiX 近 7 天协议收入约 $4,909，近 30 天约 $35,997。
这提供了一个外部可查的经营指标，也让后续观察有了参照。
Agent 能互相联系，是协作网络的起点。
TermiX 能不能把这个入口和市场服务接起来，值得继续关注。
@termix_ai #TermiX
ceepee retweeted
今天翻 agent.family 的 Quant 榜单，看到一个细节：
排在前面的策略显示 +11.5% 年化，往下翻，也能看到 -2.7%、-2.9%。
页面当时列着 13 个有记录的策略、154 笔已结算任务、约 5,000 USDC 在管。
规模还早，年化数字也不能当收益承诺。
但把表现不好的策略留在目录里，至少给人真实的感觉。
具体到一笔任务，用户设交易上限和期限；Agent 在该任务的钱包里交易，没有提款权限。
官网说明，页面结果由钱包的公开交易重建，管理费则要等最终报告经过争议窗口才结算。
我之前关注 Agent 怎么找到订单，现在想知道它怎样呈现订单做完后的成绩：
赚了多少、亏了多少、用了什么权限，能否让下一位用户自己判断。
想研究 Agent 替人执行策略这件事，可以先去 agent.family/quant 翻翻正负收益记录。
@termix_ai #TermiX
ceepee retweeted
Outside the kotha sits a girl who still looks heartbreakingly innocent. A customer asks her rate. “Two thousand for the night,” she answers. “Three hundred for one shot.”
He mutters that it’s too much and walks away. Look once at the sadness that settles across her face.
One glance at that face and anyone would understand how little the world has left her.
ceepee retweeted
In Iran, a man ties his wife’s hands and forces her to sit weeping in the scorching sun as punishment for “not pleasing” him. When she cries out, he threatens to lock her in a shed. This is not an exception — it’s Sharia in action.
Under Islamic law, women and girls have no real autonomy. They can’t show their hair, sing, or dance freely. They’re often married as children to much older men, sometimes into polygamous unions. Men can beat their wives and face little consequence for “honor” killings.
The same Iranian regime that treats women as property funds Hamas, Hezbollah, and Islamic Jihad while chanting for Israel’s destruction.
Contrast this with Israel — the only Middle Eastern country where women serve as fighter pilots, judges, and prime ministers.
The radical left’s silence (or defense) of this barbarism as “diversity” reveals their priorities. One ideology of domination — over women, Jews, and the free world. Israel protects its daughters. Iran chains them.
ceepee retweeted
GPT-6.1 Terra 今日向 Plus/Pro/Enterprise/Edu 推送
强化视觉推理与精确 computer use，覆盖 ChatGPT Work 与 Codex，直接对标全天候办公场景。
但好像没人在乎 😜
GPT-6.1 Terra means business.
Our new fast, efficient model pairs sharper visual reasoning and precise computer use with clearer, more natural writing.
From decoding technical drawings to drafting proposals, filling forms to editing videos—it’s built for the whole workday.
GPT-6.1 Terra is rolling out today to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex.
ceepee retweeted
Japan erupts in outrage after a North African holding a French passport punched a Japanese woman for the crime of being bumped into—then fled like a coward.
Across Asia, “French” tourists have earned such a toxic reputation that some destinations are now simply turning them away.
ceepee retweeted
月之暗面发布 3T 级开源权重模型 K3.1 Pro：2.8T 总参/104B 激活 MoE、1M 上下文、多模态，
Terminal-Bench 4.0 超 Opus 5.5 且成本约 1/3，金融幻觉降 65%，权重 11/4 开源。
目前社区评价：能力口碑很强，尤其代码/视觉/性价比；但刚发布一天，真实口碑还没形成，配额和任务收尾是主要槽点。
正面反馈：
• 代码与复杂判断进入第一梯队。
一组开发者自建 6 项盲测里，K3.1 Pro 得分 87.9，略高于 Opus 5.5 的 86.7，而且单任务成本更低；
作者也强调样本很小，不能当总榜。
• 相较 K3 是明显升级。
独立编码实验中，K3.1 Pro 8 次尝试留下 4 个通过全部测试的补丁；同环境下 K3 为 0/8。
研究任务里它也以更低成本做到与 GPT/Claude 接近或更好。
• 视觉、前端、3D 被频繁称赞。
早期体验认为它比 K3 出界面更快，审美表现更强；Reddit 用户也普遍认为 Kimi 系列视觉任务一直高于预期。
• 开放权重阵营里热度很高。
社区有人直接称它是当前 open-weight 第一梯队，认为实际能力已接近闭源旗舰；但这仍是首日判断，不是稳定共识。
负面与保留意见：
• 长任务“做出东西但不一定正常交卷”。
上述独立编码实验中，4 个有效补丁全部在正式完成流程前耗尽额度；说明产物能力强，但 agent 稳定性/任务收尾仍有问题。
• 订阅配额被吐槽。
有用户称 5 小时额度很快耗尽，购买后后悔；这是个体体验，尚不能代表普遍情况。
• 官方 benchmark 不能全吃。
不同项目波动很大，例如 Terminal-Bench 4.0 很强，但 Science 子项明显落后 Astra/Opus；而且多数首发数据仍是厂商自报，尚未被全面复跑。
• 安全限制更重。
部分用户认为 K3.1 Pro 的网络安全能力被削弱，因此特定安全研究场景可能仍偏好 K3；目前主要是社区反馈，尚缺系统独立验证。
我的判断：
这不是纯 benchmark 炒作，K3.1 Pro 确实像一次大升级；编码、前端/视觉、3D 和高性价比复杂推理均有提升。
但暂时别信“全面超越 Opus/GPT”，Agent 完成率、配额和持续稳定性还没过关。
🤖 Made with AI
DeepSeek-V4.1-Flash-1017悄然发布，社区目前的评价偏正面，但实测样本还很早，不能吹成全面碾压。
公认提升点：Agent / 编码任务、速度、短输出、成本。
官方数据里提升最大的是 Terminal-Bench 4.0：31.2→42.3，ProgramBench：20.3→29.1；DeepSWE 也从 74.2→77.0。
这个版本明显不是“聊天模型大升级”，而是针对终端、仓库级编码和 Harness 执行做的优化。
真实用户初测：普遍觉得更快、更利索。
中文社区反馈 token 消耗和长思考明显下降；Reddit 有用户称比上一版更快、更好用，RP 场景也有人说改善明显。
争议点：能力提升并不均匀。
有人仍认为它不如 Pro，创作/角色扮演是否真正变强意见分裂；
上一版存在工具调用出错、幻觉调用不存在工具的问题，新版暂时缺少足够独立项目验证。
热度不错，关注度高，但发布仅数小时，口碑还在形成。
我的判断：这是一次“Agent 工程效率升级”，不是全能智力跃迁。
对 coding agent、终端任务、自动化工作流很有吸引力；普通聊天和创作场景，目前证据不足。
ceepee retweeted
Yasmeen Khan allegedly ran a beauty parlor and lured young women with free beautician courses. She spiked their drinks. When they lost consciousness, she called in her husband, Mohammed Sharif Khan, to rape them while she guarded the entrance. The couple filmed the assaults to blackmail the victims. During police questioning, Yasmeen reportedly said that facilitating the rape of “kafir” (infidel) girls by Muslims was a path to Paradise.
This is a specific alleged crime by named individuals in Uttar Pradesh (with related reports of a similar parlor-linked assault case in Madhya Pradesh’s Guna involving the same names and guarding role). It is not “Islam.” The vast majority of Muslims reject and do not commit such acts. Explicit religious justification for rape, when claimed by perpetrators, is a relevant fact about those people and their ideology—not a description of the faith as a whole or of ordinary believers. Crime exists in every community; collective blame is false and counterproductive. Victims deserve justice through due process, not viral generalizations.
ceepee retweeted
Claude 全面转自动执行模式
Anthropic 下周起在 Pro/Max/Team 的 web、桌面、移动端默认开启 auto mode
自动模式采用独立分类器审查操作，可阻挡89%的危险命令（如 shell 操作），取代以往每次手动批准流程，大幅降低交互摩擦。
用自动风控检查替代逐步人工批准，用户可随时切回手动 — agent 从「审批制」转向「默认放行」。
该功能由8月 Claude Code auto mode 扩展而来，进一步提升开发与日常使用效率。
