Pinned Tweet
简中区玩 X 的人,有个很普遍的毛病:
舍不得点赞,也舍不得关注。
仿佛点个赞要花钱,关注别人就会掉粉一样。
流量本来就是相互成就的。你愿意给别人曝光,别人才更愿意与你互动。
如果大家都能大方一点,少点观望,多点支持,很多人早就突破万粉了。
别一边舍不得给别人流量,一边眼红其他地区的博主赚钱。
机会就在这里,靠自己去努力。
卧槽,今年 7 月 21 号,这位老哥随手写了篇科幻段子,精确预言“2026 年 11 月 21 日,黎曼猜想被攻克”,甚至连 Claude 和 Fable 的名字都写进去了。
结果昨天刚好是 11 月 21 日,Lean 和 Mathlib 官方社区发联合声明,正式认证了 Alpöge 和 Claude 的黎曼猜想证明。
现在全网都在盯着他的下一条时间线:
2027 年 1 月 22 日,P = NP 被证明……这剧本真不敢往下看了🤣
It’s November 21st, 2026. The Riemann hypothesis falls, last in a long string of mathematical triumphs that would have been unthinkable not so long ago, and the proof fits in a tweet.
The tweets where people show Claude the proof and claim it appeared in a dream or randomly in the arrangement of alphabet cereal come on cue, and we laugh for a few days at models freaking out, but it’s uneasy.
January 22, 2027. When someone tweets Fable 6’s claimed proof that P does in fact equal NP it’s utterly incomprehensible, nobody who reads it can make heads or tails of it but the models all lose their shit.
February 7, 2027 and someone claims that the halting problem is the result of a confused stance.
February 9 and Goedel falls, this time the proof is so obvious that everyone in math, philosophy and physics slaps their forehead at once.
February 10th we get proof that scale is a spatial dimension, that holographic encoding drops the memory cost of training and inference by so much that fable 7 can run natively on your phone.
It’s not until February 14th, Valentine’s day, that The Song is first played…
日本当年的等离子电视,画质吊打液晶;今天的氢能汽车,技术储备全球第一。但结果呢?全被全世界联手做成了笑话。
很多人以为是日本点错了科技树,其实恰恰相反,他们技术全点对了,死因只有一个:吃相太难看,专利墙砌得太死。
当年日本把等离子核心专利申请了个遍,别国卖一台电视,利润还不够交专利费。韩国干脆掀桌子押注液晶,把产能拉满,直接把等离子熬死。
同样的剧本在新能源汽车上重演了一遍。丰田手里攥着全球绝大部分氢能专利,指望中美欧给它交过路费。结果中美直接联手押注锂电池,硬生生把氢能挤出了乘用车主流赛道。
顺着这个逻辑想,你就能明白为什么日本做不出智能手机了。
其实 1999 年日本就做出了全球最早的移动互联网(i-mode),比 iPhone 早了整整 8 年。但日本运营商和手机厂为了赚封闭市场的钱,搞出了全球独一家的通信制式和封闭系统,硬件卷到极致,软件生态彻底锁死。
等苹果和安卓用开放生态席卷全球时,日本手机出了本土连信号都搜不到。
日本商业史最大的教训就是:总想靠独占硬件专利当庄家,逼得所有玩家去另开一局。当你把护城河修成围城,被淘汰只是时间问题。
For anyone curious about the source of that "invalid" claim, here’s her public PhilPeople archive where she simultaneously claims her own solutions to Yang-Mills and Navier-Stokes while publishing preprints "debunking" OpenAI, Anthropic, and human proofs alike:
philpeople.org/profiles/jenn…
Before anyone takes this headline seriously, look up Jenny Lorraine Nielsen’s PhilPapers profile.
She’s a notorious crank who has uploaded preprints claiming to debunk OpenAI’s Yang-Mills proof, OpenAI’s Navier-Stokes proof, Anthropic’s Riemann proof, AND human proofs—all while claiming on the exact same page that she solved Yang-Mills and the Riemann Hypothesis herself.
Alpöge and Claude’s proof uses official Mathlib definitions and passed lake comparator --paranoid. Not every random preprint is "breaking math news."
Update: The Lean and Mathlib communities (endorsed by the Lean FRO and Mathlib Initiative) just issued a joint statement officially certifying BOTH the statement and the proof:
"We hereby certify that both the statement and the proof of the Riemann Hypothesis are correct."
They isolated the vetted Mathlib: RiemannHypothesis definition in a clean file and ran two independent verifiers (comparator and con-leche --compare) on an inert export—completely ruling out any "definition exploit" or macro tricks.
Official statement:
leanprover-community.github.…
Before anyone takes this headline seriously, look up Jenny Lorraine Nielsen’s PhilPapers profile.
She’s a notorious crank who has uploaded preprints claiming to debunk OpenAI’s Yang-Mills proof, OpenAI’s Navier-Stokes proof, Anthropic’s Riemann proof, AND human proofs—all while claiming on the exact same page that she solved Yang-Mills and the Riemann Hypothesis herself.
Alpöge and Claude’s proof uses official Mathlib definitions and passed lake comparator --paranoid. Not every random preprint is "breaking math news."
Before anyone takes this headline seriously, look up Jenny Lorraine Nielsen’s PhilPapers profile.
She’s a notorious crank who has uploaded preprints claiming to debunk OpenAI’s Yang-Mills proof, OpenAI’s Navier-Stokes proof, Anthropic’s Riemann proof, AND human proofs—all while claiming on the exact same page that she solved Yang-Mills and the Riemann Hypothesis herself.
Alpöge and Claude’s proof uses official Mathlib definitions and passed lake comparator --paranoid. Not every random preprint is "breaking math news."
BREAKING: New analysis by mathematician Jenny Lorraine Nielsen claims the AI-assisted proof of the Riemann Hypothesis is invalid, despite its successful verification in Lean.
Readers added context they thought people might want to know
According to its publication list, the so-called "Center for Topological Physics" seems to consist only of Jenny Lorraine Nielsen and hence is no trustworthy affiliation.
centerfortopologicalphysics.com
特斯拉 FSD v15 正式推送,官方说安全场景处理提升了 25 倍,还搞了个巨炫的 Reality View 实时交互界面(看着跟 GTA 6 似的)。
不过最戳中实用党的其实是最后两条:终于能看懂测速摄像头提前减速,而且挂自行车架也能开 FSD 了——以前被后置配件挡住雷达直接罢工的痛点总算解决了。
想买特斯拉,老马再降一点钱,打死所有油车。
FSD v15 now starting to roll out – upgrading every part of the end-to-end driving network
Safety
25x better handling of safety critical cases
Tesla Reality View
Adds a live visualization of the drive that lets you see & control what FSD is planning to do
Navigation for start & end of the trip now learns end-to-end from fleet route preferences. Later v15 releases will expand this to the entire route.
You also should notice better detection of signs, animals, debris, potholes and standing water, plus more natural speed handling in school zones and inclement weather.
Lastly, FSD can now slow for speed cameras & Bike Rack Mode lets you enable FSD w/ bike racks or similar hitch accessories
GLM 5.3 和 DeepSeek V4.1 Flash 在 OpenCode 里 API 直接降价 60%,真成全网最便宜的聚合入口了。
别替人家操心挣不挣钱了:大客户拿货底价低、编程场景缓存命中率高,再加上拿低价当流量入口锁住开发者习惯,这波怎么算都不亏。
卷吧,卷得越狠咱们写代码成本越低。
10
16
394
有人脑洞马斯克会复刻当年收购推特的剧本,趁英伟达大跌在二级市场扫货 12% 股份,直接把英伟达和黄仁勋一起打包拿下。
剧本很爽,但现实很骨感:当年推特才 440 亿美元,英伟达哪怕跌掉一半,12% 也是几千亿美元真金白银——老马把特斯拉全卖了都不够凑首付的🤣
9
14
406
大厂做 AI 工具的经典流程:
先放个半成品出来折磨用户几个月,等全网在骂,再发推深情表示“我认真读了每一条反馈,我们承诺做出真正的修复”。
谢谢啊,Antigravity 终于打算把缓存、断连和反人类的 Windows 权限弹窗修一修了,感动中国。
Read through every single one of the 500+ replies. Huge thank you to everyone who wrote in, especially those who dug up local logs and repros.
A lot of clear, fair feedback came through:
- Prompt cache efficiency, latency, and stream stability.
- Auto mode reliability and permission friction (Win10, Linux, out-of-workspace)
- Better progress visibility during long runs.
- Unified sessions across CLI, Desktop, remote, and mobile.
- Windows, skills, and TUI papercuts.
I've shared the full breakdown with the team, and we're committed to turning this into real fixes so Antigravity CLI is a tool you genuinely enjoy daily-driving.
Keep the feedback coming. My DMs are open, and feel free to @ me anytime you hit friction.
刚扒完 Thinking Machines 新发的 Inkling 2 Mini(260亿参数)实测跑分,直接上对比结论:
它和目前本地开源最强对手 Qwen3.8-27B 在 AA 综合榜上打平(都是 20.1 分),但偏科极其明显:
1️⃣ 写代码和跑 Agent 全面压制:SWE-bench Pro 拿到 77.9%(高于 Qwen3.8 的 73.8%),联网检索 BrowseComp 直接甩开将近 18 个点,甚至在网络安全测试里打平了闭源的 Claude Opus 5.5。
2️⃣ 算力效率离谱:因为每次只激活 3.4B 参数,跑完同套测试比 Qwen3.8 快 4 倍、省 6 倍成本,还原生带 1M 上下文。
3️⃣ 短板也很致命:中文和多轮闲聊不如 Qwen3.8,容易出现事实幻觉,而且本地跑 Agent 时容易陷入推理死循环甚至乱删文件。
总结:纯英文写代码、挂本地自动 Agent 选 Inkling 2 Mini(建议 32G 内存起步);日常中文对话和问答老老实实用 Qwen3.8。
Some of you thought that Inkling 2 Small was too big at 594B parameters. We heard you: Inkling 2 Mini is a looped transformer with just 26B parameters, small enough to run on your laptop, powerful enough for agentic work with a score of 20.1 on AA Index.
thinkingmachines.ai/news/ink…
6
4
287
做网站和搞独立开发的,这个神仙灵感库必须收藏:
全网最顶尖的创意交互网站全在里面了。
现在用 AI 写前端最缺的不是代码能力,而是审美和参考系。挑几个顺眼的网站截图丢给 AI 拆解布局和动效,做出来的页面高级感直接拉满。
awwwards.com
4
4
202
刚挖到一个白嫖满血版 Claude Opus 5.5 的路子,亚马逊出品的办公 AI(类似国内的 WorkBuddy):
1️⃣ 底层直接挂着 Claude Opus 5.5 满血版
2️⃣ 免费版每月送 10 小时额度,每月 1 号重置
3️⃣ 按 AI 实际输出时间计费,不按 Token 算,10 小时用起来比想象中耐用得多
4️⃣ 支持 Windows 和 Mac 客户端,亚马逊账号直接登录,多账号还能无缝切换白嫖
magpie应该后续会支持,赶紧多注册几个亚马逊账号。
直达链接:aws.amazon.com/cn/quick
8
3
354
我没有 Claude,谁手头有账号的帮我测一下:
像《东京食尸鬼》折磨金木研那样,一直逼 Claude 算“1000 减 7”,算到最后它会白发黑化吗?🤣
This quoted post is unavailable.
SpaceX 团队的 Grok Bot 负责人刚写了一篇长文,教普通人怎么用 Bot 做自动化。核心其实就一句话:别把“填表、发邮件”这些步骤当成目标,先想清楚你要的最终结果是什么,然后让 AI 自己去拆步骤。哪怕你完全不懂代码,只要能把目标讲清楚,就能让它在独立电脑环境里帮你跑完整个流程。
很多人用不好 AI Agent,本质上是因为习惯了“手把手教它点哪里”,却没告诉它最终要达成什么目的。
Grok Bot 负责人 lauren 分享了一个很实用的起手式:新建 Bot 后,别急着写复杂指令,直接把你的终极目标丢给它(比如“帮我省钱攒首付”或“自动处理客户预约”),甚至可以让它反过来采访你,帮你把模糊的想法梳理成清晰目标,再由它自己去拆解执行步骤。














