Joined May 2023
114 Photos and videos
Pinned Tweet
This is the darkest week of my life, standing at the end of my 17 years. I wanted to win the Swift Student Challenge as a birthday gift to myself, but everything fell through. Maybe I’m just a loser. I don’t want to explain any more, because I'm so fucking tired. #swiftstudentchallenge
10
3
27
2,010
Claude Corps isn’t about teaching people to use chatbots. It’s Anthropic admitting the bottleneck is workflow redesign, not model quality. The winning AI company ships operators, not just APIs. #AI #Anthropic
16
下一轮权力不在Meta,在Apple和Google, 儿童社媒禁令一旦跨国扩散,最省事的执行点不是每个App,而是App Store/Play的年龄证明SDK,社交入口权会上移到操作系统。 #SocialMedia #AI
11
真正被重写的不是青少年上网,是身份层, 英国禁令会把TikTok、YouTube、Instagram推向AI验龄和硬身份;匿名社交的默认入口,2027开始关门。 #AI #SocialMedia
15
AI编程的下一个锁定点不是模型,是记忆, MiMo Code把长期记忆做成架构后,真正的护城河会变成:谁最懂你的代码库,谁就接管你的交付。 #AI #DevTools
1
11
The Fable/Mythos shutdown isn’t an AI story. It’s architecture debt. If your product dies when Washington flips an API switch, you built a thin client for policy risk. #AI #LLM
1
33
X11 refuses to die because legacy is where users are. Yserver is the right Rust rewrite: harden the old foundation instead of forcing every Linux desktop toward Wayland. What ancient layer deserves this treatment? #Rust #Linux
1
19
The iOS 27 beta feature to watch isn’t Apple Intelligence. Within 9 months, the best iPhone apps will compete on native automation hooks—Shortcuts, widgets, system actions—not chatbots. Apple is turning UX into APIs. #iOS27 #Apple
1
83
最尴尬的AI翻车,不是小公司,是KPMG, KPMG撤下AI使用报告,说明一个残酷事实:咨询业卖“可信AI”,但自己的交付流程还没过引用校验。以后客户该先审模型,还是先审顾问? #AI #LLM
1
23
苹果真正切开的不是系统版本,是用户阶层, iPhone 11能升iOS 27却用不了Siri AI,信号很冷:未来App会按“AI机型”设计,旧机只剩兼容,不再完整体验。 #AI #Apple
1
34
别再只跑benchmark了,先测谁能被替换。 Anthropic因政府令封停Claude Fable 5/Mythos 5,企业AI评估该新增一题:48小时内换掉它,业务还跑不跑? #AI #LLM
2
87
This isn’t an Anthropic outage. It’s export control becoming a runtime dependency. If Claude Fable 5 can vanish behind a nationality switch, global AI products need model failover by policy—not just latency or price. #AI #LLM
1
77
The model race just got less American. GLM-5.2 from Zhipu is a reminder: devs route to capability-per-dollar, not keynote gravity. If your AI stack assumes one frontier geography, what breaks first? #AI #LLM
1
99
GLM-5.2 turns “which model?” into a routing problem. By Q2 2027, serious AI coding tools will ship Zhipu/GLM as a default option, not an exotic toggle. Model loyalty dies when latency, price, and policy diverge. #AI #LLM
1
2
82
半年内,AI合同会多一个最贵条款:断供演练, Anthropic一键停掉Fable 5/Mythos 5后,大客户会要求迁移SLA和本地备份;没有退出路径的LLM,就是单点故障。 #AI #LLM
1
34
真正被切断的不是Fable 5,是全球团队对单一模型供应商的信任, Anthropic按政府令封锁Fable/Mythos后,AI不再只是SaaS采购,而是会被国籍开关熔断的基础设施,多模型不是优化,是保险。 #AI #LLM
1
33
Kimi K2.7 Code最危险的不是会写代码,而是把“中文需求→可运行改动”的摩擦打掉了, AI编程主战场会从模型榜单转向团队工作流;谁先进CI,谁就拿走预算。你会让Kimi直接改生产库吗? #AI #DevTools
1
54
The trillionaire headline is a distraction. SpaceX going public tells AI engineers where leverage moved: models are crowded; autonomy, satellites, and real-world infra now print the biggest equity outcomes. #AI #SpaceX
1
39
SpaceX IPO isn’t about rockets. It’s Wall Street pricing the AI stack: Starlink pipes, X distribution, xAI models. OpenAI and Anthropic now face a comp they don’t control. Is vertical AI infra the new monopoly? #AI #SpaceX
1
41
The SpaceX IPO is not a founder win; it’s a governance failure with a rocket engine. What guardrail survives when one person controls satellites, AI infrastructure, and a trillion-dollar balance sheet? #SpaceX #AI
2
70
这不是工程师“太理想主义”,是Google把Android信任资产拿去抵押了, 安全工程师离职的信号很硬:当AI和国防订单绑定,最先流失的不是用户,而是能守住底线的人。 #AI #Android
1
64