Slow takes on fast markets. First-order thinkers react to news — I trade the second-order effects. Compounding curiosity in tech, crypto, macro.

USA
数据没多,场景多了。 开源手机智能体 OpenMobile-2, 训练场里 110 多个 App。 陌生 App 上, 成功率从 60.7% 提到 67.5%。 再叠强化学习, 总分反而从 80.8 掉到 79.9。
5
我回朋友消息, 一直用 AI 帮我改语气。 今天发出去,他回一句: 「你说话怎么这么客气。」 改得越像人, 越不像我。
1
29
AI-assisted research found two PHP flaws behind 70% of the web. In the same program, 1,316 of 1,995 findings were still unpatched. The bottleneck moved to whoever has to read them.
23
我设了 6 个闹钟, 只用得上最后一个。 从 7:10 到 7:18, 中间按掉 4 次, 第 5 次把隔壁吵醒了。 最后只留 7:18。 多设几个闹钟 不会让人早醒, 只会让人多赖 8 分钟。
2
1
233
一种气体,进了关键材料清单。 23 日,美国能源部 把氦气列为第 20 种。 晶圆冷却、光刻、检漏, 芯片厂离不了它。 全球需求增速已超产量增速, 美国约占全球产量 43%。 清单上的名字, 比订单更早决定扩产。
3
262
Meta’s Q3 margin story needs two lenses. FACT: Revenue rose 26%; ad impressions rose 10% and price per ad 14%. Reported operating income fell 37%, including a $9.89B legal charge. MY READ: Excluding that charge, operating income was $22.88B—about a 35% margin versus 40% a year ago. The ad engine is growing, yet the adjusted margin is still tighter. Free cash flow was -$2.12B, with capex (including finance leases) at $34.9B. The investment question: can ad monetization keep funding AI and infrastructure at this pace? What matters more to you here—ad growth or cash returns on AI spend?
2
1
59
The AI race may be decided by more than who builds the best model. AP reports that countries including Kazakhstan are trying to benefit from both U.S.- and China-linked AI ecosystems. Kazakhstan brings two less obvious assets to the story: critical minerals and relatively cheap power, alongside plans for a major data-center hub. My investing takeaway: follow the bottlenecks, not just the headlines—electricity, land, cooling, minerals, and the rules governing where compute can be built. Which layer do you think investors overlook most: chips, power, or data centers?
3
2
75
性价比的选择
你们在 WorkBuddy 里最常用哪个模型? 我把 17 款模型按 性价比从高到低 做了张图。 这次不只是看性能, 还把 积分消耗 一起放进去了。 有些模型属于: 能力是强,但真不便宜。 有些则是: 积分花得少,干活还挺猛。 如果你也在用 WorkBuddy, 欢迎看看这张图,顺便说说你现在的主力模型是哪个? #AI工具 #WorkBuddy
1
3
105
这周看美股科技股,最容易犯的错:把同一天的波动全归因于财报。 事实:美联储 10 月 27–28 日开会;微软定于 28 日收盘后公布 FY2027 第一季度业绩;亚马逊定于 29 日举行第三季度业绩电话会。 我的观察:若股价大幅波动,先拆成两问:利率预期变了吗?公司的增长、利润率或资本开支指引变了吗?时间重合,不等于原因相同。 你看大型科技股财报,最先关注收入增长、利润率,还是 AI 投入回报?
6
7
265
各位领取了 Google ai pro 的有没有收到重新验证学生资格的邮件啊 怎么搞啊
1
2
163
今年万圣节,糖果也开始“缩量”了吗? AP 援引 NielsenIQ 数据:截至 10 月 17 日的一个月,美国万圣节糖果销售同比下降 6.7%;目前只有小包装还在增长。 这不只是节日消费新闻。看消费股时,我会追问:顾客少买了,还是改买更小、更便宜的包装?前者可能意味着需求变弱,后者则会影响客单价和产品组合。 你最近买糖果时,最明显的变化是价格、包装,还是干脆少买?
3
53
拼单群待了 2 年, 买过 31 单, 自己用上的 4 件。 剩下 27 件, 都是替别人凑的。 我从来不主动买, 但每次都第一个接。 凑单的人最难说不。
4
2
135
眼镜腿松了, 我自己拧了 4 次螺丝。 第一次撑 3 天, 第二次撑 5 天, 第三次拧完当场又滑下去。 昨天去眼镜店, 师傅拧了 1 分钟, 说螺丝孔已经磨圆了。 螺丝还能拧紧, 孔接不住了。
1
32
AI Storage Is Booming. So Why Isn’t a Strong Earnings Beat Enough? Western Digital just delivered a powerful quarter: • Revenue: $4.38B, up 55% YoY • Adjusted EPS: $4.64, up 161% YoY • Next-quarter revenue guidance: $5.0B–$5.2B, above analyst expectations of roughly $4.57B • Expected adjusted gross margin: 60%–61%, up from 58.1% Yet WDC shares barely rose in Friday trading, while the Roundhill Memory ETF gained about 4%. The market is asking a harder question: how much future growth is already priced in? citeturn883530news15turn883530search8 Here’s the bigger AI infrastructure thesis: AI creates a storage problem, not just a computing problem. Training generates enormous datasets; inference, video, and AI agents create more data that must be retained and accessed. HDDs still have a role. High-capacity hard drives can store large volumes of data at a lower cost per terabyte than flash storage, making them relevant to hyperscale data centers. Pricing power is the key variable. If demand continues to exceed supply, higher prices and better margins can drive earnings growth. But if competitors add capacity or customers push back on pricing, today’s margins could prove cyclical. Expectations matter. When a stock has already roughly doubled in a year, beating estimates may not be enough. Investors want evidence that the next few quarters will be even better. The bull case is that AI’s data footprint keeps expanding. The risk is that investors mistake a strong part of the cycle for permanent growth. Would you buy storage stocks on the AI demand thesis—or wait for valuations to cool? #AI #WesternDigital #WDC #Semiconductors #Investing
2
203
AI 存储需求很强,为什么业绩超预期还不够? 西部数据(WDC)交出了一份相当亮眼的财报: • 营收:43.8 亿美元,同比增长 55% • 调整后每股收益:4.64 美元,同比增长 161% • 下一季度营收指引:50 亿—52 亿美元,高于分析师约 45.7 亿美元的预期 • 预计调整后毛利率:60%—61%,高于上一季度的 58.1% citeturn883530search8 但在这轮反弹中,存储板块表现强于 WDC 本身:文章报道时,Roundhill Memory ETF 上涨约 4%,而 WDC 股价仅上涨约 0.5%。这反映出市场关注点正在变化:不只是业绩好不好,而是未来还能不能持续超预期。 citeturn883530news15 这背后有几个值得关注的投资逻辑: AI 不只是算力问题,也是数据存储问题。 模型训练需要海量数据,推理、视频和 AI Agent 的普及,也会持续增加数据保存与调用需求。 机械硬盘仍有独特价值。 对超大规模数据中心来说,高容量 HDD 在每 TB 成本上仍有优势,适合存放海量数据;闪存则更适合强调速度的场景。 真正决定盈利弹性的是定价权。 如果需求持续超过供给,价格和利润率就有机会提高;但如果竞争对手扩产、客户议价增强,利润率也可能回落。 估值决定股价怎么走。 当一只股票一年内已经大幅上涨,单季业绩超预期可能早已被市场计入。投资者想看到的,是接下来几个季度还能继续超预期。 看多逻辑是:AI 产生和处理的数据越来越多,存储需求有望持续扩张。主要风险则是:市场把一个强劲的行业周期,当成了永远持续的高增长。 你会基于 AI 存储需求买入相关股票,还是等估值回落后再考虑? #人工智能 #西部数据 #WDC #半导体 #投资
7
5
460
实时视频的问题, 是像了却改不动。 20 日 StreamEval 测了 37 个模型、300 道题。 参考图编辑「完整成功」最高分: 15.00。 披风换金缎再取消, 屏幕上还是那件灰的。 看起来像真的, 和照着做,是两件事。
21
My fastest reply goes to people I like least. The group chat gets 4 minutes. The one I actually want to talk to gets 9 drafts and 3 days. Three days I chose. Four minutes I didn't.
18
昨夜布伦特原油涨 3.14%, 报 116.72 美元, 创 7 个月新高。 同夜斯伦贝谢涨 7.06%, 美国航空跌 3.89%。 一桶油涨的钱, 从下游挪到了上游。
5
4
277