共收录 22 篇深度内容
📌 今日概览
- 开源模型与 AI 代理继续下沉到个人设备与日常工作，离线运行、免费剪辑工具、家具设计等用例密集出现。
- Claude Code 成为今日最集中的工具符号，涵盖建站、视频剪辑、家具设计、iOS 开发与维权代理。
- 数据中心叙事出现两面：耗水被证伪为小比例，而内蒙古风电园区以 800MW 规模快速建成。
- AI 能力争论聚焦方法论与核查：Chollet 强调方法泛化，Mollick 称 AI 核查已超专家。
🔥 必读 Top3
- Chubby♨️ 数据中心耗水是个错误说法：2024 年美国大型数据中心现场耗水约 190–230 亿加仑，约为同年美国高尔夫球场 5,320 亿加仑灌溉用水的 4%。
入选理由：用美国高尔夫球场灌溉用水作对照，给出数据中心耗水的可核查数字，纠正常见误读。 - SemiAnalysis 作者称实地探访中国内蒙古的“Stargate”：一座风电供能的 800MW 园区，一年内建成 120MW，并有 5MW SST 原生 800V 直流试点与无中央冷水机组的冷却设计。
入选理由：实地探访内蒙古风电园区，披露 800MW 规模、一年建成 120MW 及 800V 直流试点等一手细节。 - Aakash Gupta 犹他州派卧底探员在 Kalshi 下注约 115 美元，Kalshi 两位创始人因此被控 13 项重罪，最高一项可判 15 年；指控公布次日投资者向 Kalshi 注资 10 亿美元。
入选理由：卧底下注、13 项重罪指控与次日 10 亿美元注资构成完整事件链，信息密度高。
🧭 主题分组
- 开源模型（3 条）—— 开源模型向个人离线运行与免费工具化推进，Claude Code 成为快速造应用的入口。
- AI 代理（4 条）—— 代理从建站、家具设计延伸到消费维权与 iOS 开发，安全扫描也开始由机器人承担。
- 数据中心（2 条）—— 耗水争议被数据澄清，同时中国风电园区的建设速度与供电架构成为新观察点。
- AI 判断（5 条）—— 围绕 AI 能力边界形成两派：核查与研究效率被看好，方法泛化仍被强调为关键。
- 商业与争议（4 条）—— 预测市场遭执法与资本同时加注，AI 公司增速受质疑，开源项目新增限制引发反弹。
- 产业与生活（4 条）—— 半导体史、3D 重建、开源硬件与育儿感悟并置，呈现技术背后的长期积累与日常视角。
📖 详细内容
开源模型
开源模型向个人离线运行与免费工具化推进，Claude Code 成为快速造应用的入口。
Alex Veremeyenko
｜ @alex_verem
｜ 影响力: 10.4万粉丝 · 互动分 10639.0
Alex Veremeyenko
｜ @alex_verem
｜ 影响力: 10.4万粉丝 · 互动分 10639.0
💡 核心观点: 呼吁本周末下载开源 AI 模型：单个文件 5–20GB，可在自有笔记本离线运行，无需账号、订阅或联网，约 20 分钟即可装好。
可信度: 60/100
事实核查:
- 无法核实: 开源模型为单个文件，大小 5 到 20GB，可在现有笔记本上运行（属概括性描述，未指明具体模型与硬件要求）
- 无法核实: 模型文件在 2036 年给出的答案与今天相同（为作者推断，原文无验证依据）
原文内容:
I am begging you, download an open source AI model this weekend. It's one file, 5 to 20GB, and it runs on the laptop you already own. You won't need an account, a subscription or an internet connection once it's on your drive. Your paid AI can change overnight, cost more next year or lock you out. A model file you keep gives the same answers in 2036 that it gives today. Grab Qwen or Gemma, install a free open-source app to run it, and ask it the questions you'd never type into a website. Setup takes about 20 minutes. You'll be glad it's done the day your account gets locked.
⏰ 11:01 | ❤️ 8473 点赞 | 📝 114词 | 入选: 高赞 8473 | 查看原文 →
Andrej Karpathy
｜ @karpathy
｜ 影响力: 443.5万粉丝 · 互动分 8261.0
Andrej Karpathy
｜ @karpathy
｜ 影响力: 443.5万粉丝 · 互动分 8261.0
💡 核心观点: 引用 Tristan 的 YouTube 频道榜单，用 Claude Code 三分钟做出 TristanTube，汇集 4.7 万条硬核视频并提供电视模式。
可信度: 70/100
事实核查:
- 原文支持: TristanTube 收录 4.7 万条视频（原文给出数字，无独立核验）
- 无法核实: 用 Claude Code 三分钟一次做成（作者自述效率，无可核验细节）
原文内容:
Earlier this week Tristan published his top YouTube channels: https://t.co/NZ3UCNo6cT Which looked excellent so clearly I had to turn it into TristanTube: https://t.co/l0KyW1WdNH 47K super nerdy videos +TV mode for laid back watching. (One-shotted with Claude Code in 3 minutes)
引用 @karpathy: Weekend hack: HackerTube. I was annoyed with YouTube's recommendations so I had Claude filter down 43M comments over 20 years of HN to 6,454 videos that keep being recommended. *Significantly* better viewing experience, ~2 hours with Opus 5.5:
https://t.co/JIhXN7OSh1
– Excellent recs without any lo…
⏰ 17:07 | ❤️ 6217 点赞 | 📝 43词 | 入选: 高赞 6217 / 引用讨论 | 查看原文 →
Riley Brown
｜ @rileybrown
｜ 影响力: 26.4万粉丝 · 互动分 5163.0
Riley Brown
｜ @rileybrown
｜ 影响力: 26.4万粉丝 · 互动分 5163.0
💡 核心观点: 预告下周发布一款免费开源视频剪辑工具，可直接在 Claude Code、Codex 和 grokbot 内使用，评论该帖即可获取访问。
可信度: 25/100
事实核查:
- 无法核实: 该工具 100% 免费且开源（尚在预告阶段，原文未给出项目名称或仓库）
原文内容:
if u want access to what i believe is the best free video editing tool that's releasing next week comment on this post. Works directly inside claude code, codex, and grokbot. again 100% free and open source.
⏰ 07:09 | ❤️ 2326 点赞 | 📝 37词 | 入选: 高赞 2326 | 查看原文 →
AI 代理
代理从建站、家具设计延伸到消费维权与 iOS 开发，安全扫描也开始由机器人承担。
@levelsio
｜ @levelsio
｜ 影响力: 99.0万粉丝 · 互动分 2647.5
@levelsio
｜ @levelsio
｜ 影响力: 99.0万粉丝 · 互动分 2647.5
💡 核心观点: Claude Code 能逐件设计家具并给出材料清单，让用户自行下单组装，价格远低于大品牌。
可信度: 62/100
事实核查:
- 无法核实: 自行设计家具、直接采购原材料，可比 Poliform 等高端品牌省 85%（基于作者自述案例与推算，无第三方数据支持）
- 无法核实: 一张 30,000 美元的沙发可降到 4,500 美元（按 85% 降幅算术推得，原文无实际采购数据）
原文内容:
Claude Code can design furniture part by part and then give you a bill of materials to order each part and put it together yourself And way cheaper than big interior brands sell you their furniture Furnishing my house I discovered how absolutely insane prices are for the most basic cheap materials put together with a fancy brand tied to it One shop had a sofa for $30,000 😂 I think with Opus 5.5 and other models easily operating 3d programs like Blender and FreeCAD (both free) it makes more sense to cut out the middle man and go straight to suppliers and design and build your own furniture Which in a way makes customers more powerful (they can get exactly what they want and usually for cheaper without the markup) Anyway @theannalux designed this table from scratch, she's not a furniture designer, not a coder, she just had an idea of a style and worked with Claude Code in the terminal (far superior to the app I will keep saying this) If you design your own furniture and buy the raw materials directly, you can save 85% compared to premium brands like Poliform That means the $30,000 sofa becomes $4,500! And the $8,000 dinner table becomes just $1,200! You can see where the idea comes from that AI will make everything 5x to 20x cheaper, and it's already starting!
⏰ 18:17 | ❤️ 1823 点赞 | 📝 221词 | 入选: 高赞 1823 / 长文 | 查看原文 →
Marc Lou
｜ @marclou
｜ 影响力: 42.3万粉丝 · 互动分 1474.5
Marc Lou
｜ @marclou
｜ 影响力: 42.3万粉丝 · 互动分 1474.5
💡 核心观点: 作者讲述用 AI 代理处理坏鞋维权：代理向门店、区域及全球客服升级并准备支付争议材料，最终拿到 850 AED 退款；他并提醒此类操作易被滥用。
可信度: 70/100
事实核查:
- 无法核实: bought a pair of VIVO Barefoot shoes in Abu Dhabi 2 weeks ago for 850 AED (~$230)（金额、地点、品牌与时间均为作者自述，原文未展示票据或截图。）
- 无法核实: Today, I checked my bank and noticed 850 AED had landed in my account（退款到账为作者个人陈述，原文未提供银行记录。）
- 无法核实: The store refused international exchange. The agent pushed for a remote refund（流程细节来自对代理聊天记录的转述，原文未展示原始对话。）
原文内容:
AI is so, so good. I bought a pair of VIVO Barefoot shoes in Abu Dhabi 2 weeks ago for 850 AED (~$230). I absolutely loved wearing them (you should try them!). But 2 days later, the back of one shoe broke. Instant love-hate relationship. So I spun up an AI agent, gave it the receipt and photos, and asked it to get me a replacement delivered to my door. Or a refund if that wasn’t possible. Life went on. I forgot about it. Today, I checked my bank and noticed 850 AED had landed in my account. I opened the agent’s chat to see what happened: 1. The agent requested a replacement delivered to my door. 2. The store had stock, but Cyprus delivery stalled. 3. The agent escalated to regional and global support. 4. The store refused international exchange. The agent pushed for a remote refund. 5. The requested reply deadline passed. No answer. 6. The agent prepared Wise dispute evidence as backup. The store eventually processed the refund. 7. The money landed. It’s crazy that I can have an army of superintelligent agents fighting for me, 24/7, while I get on with my life. The downside: a lot of people are going to abuse this. Filing a claim has become so easy.
⏰ 13:13 | ❤️ 1038 点赞 | 📝 206词 | 入选: 高赞 1038 / 长文 | 查看原文 →
GREG ISENBERG
｜ @gregisenberg
｜ 影响力: 75.9万粉丝 · 互动分 1423.0
GREG ISENBERG
｜ @gregisenberg
｜ 影响力: 75.9万粉丝 · 互动分 1423.0
💡 核心观点: 作者引用「Apple 把 Xcode 放进 Claude Code」的消息，给出用 AI 代理选品、验证需求、开发与测试 iOS 应用的清单。
可信度: 65/100
事实核查:
- 无法核实: Apple just put Xcode inside Claude Code. It's an official Apple plugin, now in Claude Code's plugin directory（引自所附推文，原文未提供 Apple 官方公告链接。）
- 原文支持: filter for apps with a rating under 3.5（该筛选条件在原文中以明确指令形式给出。）
- 无法核实: apple's foundation models for on-device ai, so your inference cost is $0（原文称端侧基础模型可使推理成本为零，未给出计费依据。）
原文内容:
now that apple put xcode inside claude code, here’s how i’d find, build and grow an iOS app with AI agents: find it - open appfigures or sensor tower and sort utilities by revenue - filter for apps with a rating under 3.5 - bonus points if the top app hasn't shipped an update in 6+ months - check the keywords they rank for in appfigures. low difficulty + high revenue is the sweet spot - scroll the "you might also like" row on the top app's page. every app there is another market to check - or find a startup idea on https://t.co/a5ARFnvky2 based on social sentiment validate it - use facundoolano/app-store-scraper to pull every review for the top 20 - give them to an agent: "cluster the 1 and 2 star reviews into the top 5 complaints and quote the users" - pick the app where the complaints are about the product, not the price - if people complain about the price but keep paying, that's demand. if they complain about bugs, that's your opening - search reddit for "alternative to [app]". every thread is a list of people waiting for you build it - claude code + apple's new xcode plugin to build, preview and fix - cameroncooke/XcodeBuildMCP so the agent can build, test and run the simulator on its own - joshuayoes/ios-simulator-mcp so it can tap, swipe and screenshot like a user - write a skill with the top 5 complaints so every feature gets checked against them - apple's foundation models for on-device ai, so your inference cost is $0 - app intents so siri and personal agents can use your app too - add an "import from [competitor]" button. make switching take 10 seconds test it - maestro tests every flow the reviewers complained about - the agent runs them on every screen size before each build - if the competitor's worst flow takes 6 taps, yours takes 2 - sentry for crashes, telemetrydeck for analytics - testflight it to 20 people from those reddit threads before launch monetize it - revenuecat for subscriptions and mrr - superwall to test paywalls without shipping a new build - paywall right after the first win, never on launch - annual as the default with a 3-day trial, weekly as the anchor - turn on win-back offers for everyone who cancels rank it - your screenshots answer each complaint directly. "no ads." "works offline." "cancel anytime." - fastlane snapshot + frameit regenerates every screenshot automatically - a/b test them with app store connect's product page optimization - have the agent localize your listing into 10 languages. most competitors only rank in english - ask for a review right after a win with apple's built-in review prompt, never after an error grow it - find 10 creators in your niche on tiktok one. small monthly base + a bonus for every 100k views - give each one a fresh tiktok account that only posts about your niche - apify's tiktok scraper pulls the top 100 hooks in your niche every week. that's your hook bank - every creator tests 3 new hooks a day in capcut. anything that loses people in the first 3 seconds gets cut - arcads or heygen for ai ugc when you need 50 variations of a winner - postiz to schedule everything across tiktok, reels and shorts - faceless slideshow accounts posting 3x a day: problem, screenshot, result - when an organic video takes off, put money behind it with tiktok spark ads instead of making new ads - a unique apple offer code for every creator so you know exactly who drove each subscriber - a quiz landing page on framer that ends in "download the app." web-to-app funnels convert better than cold traffic - appsflyer to track which video, creator and ad actually turned into revenue - apple ads on the competitor's name i could do a whole episode on @startupideaspod if people are interested. i will never gatekeep anything im learning in real-time. welcome to the golden age of ios apps, because now anyone can build one
引用 @RoundtableSpace: Apple just put Xcode inside Claude Code.
It's an official Apple plugin, now in Claude Code's plugin directory:
• Build, run and test your app
• Render SwiftUI previews
• Drive simulators and real iPhones
• Read build logs and errors
• Search Apple's developer docs
Claude can now bu…
⏰ 00:45 | ❤️ 1007 点赞 | 📝 648词 | 入选: 高赞 1007 / 引用讨论 / 长文 | 查看原文 →
Marc Lou
｜ @marclou
｜ 影响力: 42.3万粉丝 · 互动分 739.04
Marc Lou
｜ @marclou
｜ 影响力: 42.3万粉丝 · 互动分 739.04
💡 核心观点: 作者称“黑色十月”后采取更多措施：移除几乎全部浏览器扩展、把仓库依赖精简到最低限度、设置 Grok 机器人每小时扫描网络，发现安全问题就提交 PR 并发送 Telegram 提醒。
可信度: 62/100
事实核查:
- 原文支持: Removed almost all my browser extensions（原文直接陈述作者的个人操作，属自述，无第三方数据佐证。）
- 无法核实: Set up a Grok bot to scout the internet every hour and file a PR if a security issue is detected (with Telegram alerts)（原文描述其搭建的自动化流程，细节具体，但文本内没有可验证的运行证据。）
原文内容:
I've taken a few more measures after Black October: - Removed almost all my browser extensions - Trimmed all my repo dependencies to the strict minimum - Set up a Grok bot to scout the internet every hour and file a PR if a security issue is detected (with Telegram alerts)
引用 @marclou: I spent the last 48 hours rotating passwords, adding 2FA everywhere, and spawning agents to audit all my startups' codebases.
I tried to brute-force my own code, but, of course, OpenAI and Anthropic models wouldn't let me.
I also ordered a YubiKey for myself and to apply to OpenAI's Daybreak progr…
⏰ 11:43 | ❤️ 450 点赞 | 📝 49词 | 入选: 高赞 450 / 引用讨论 | 查看原文 →
数据中心
耗水争议被数据澄清，同时中国风电园区的建设速度与供电架构成为新观察点。
Chubby♨️
｜ @kimmonismus
｜ 影响力: 15.7万粉丝 · 互动分 2255.0
Chubby♨️
｜ @kimmonismus
｜ 影响力: 15.7万粉丝 · 互动分 2255.0
💡 核心观点: 数据中心耗水是个错误说法：2024 年美国大型数据中心现场耗水约 190–230 亿加仑，约为同年美国高尔夫球场 5,320 亿加仑灌溉用水的 4%。
可信度: 85/100
事实核查:
- 原文支持: 2024 年大型美国数据中心现场耗水估计 190–230 亿加仑（Han et al., 2026）（给出文献与年份，但摘要未含完整引用细节）
- 原文支持: 美国高尔夫球场同年灌溉用水 5,320 亿加仑（GCSAA, 2025）（有机构名与数字，但未说明统计口径细节）
- 存在偏差: 关于数据中心的许多说法是谎言（属作者评价，原文未提供逐条反驳依据）
原文内容:
The claim that data centers are a large drain on the nation's water does not hold up. Its just false. In 2024, large U.S. data centers consumed an estimated 19–23 billion gallons on site (Han et al., 2026). For scale, that’s about 4% of the 532 billion gallons U.S. golf courses applied for irrigation that same year (GCSAA, 2025). The golf figure includes recycled water, and the data-center figure covers on-site consumption only. There are so many false statements about data centers that, given better knowledge, must be labeled as lies.
⏰ 11:19 | ❤️ 1547 点赞 | 📝 91词 | 入选: 高赞 1547 | 查看原文 →
SemiAnalysis
｜ @SemiAnalysis_
｜ 影响力: 18.3万粉丝 · 互动分 638.5
SemiAnalysis
｜ @SemiAnalysis_
｜ 影响力: 18.3万粉丝 · 互动分 638.5
💡 核心观点: 作者称实地探访中国内蒙古的“Stargate”：一座风电供能的 800MW 园区，一年内建成 120MW，并有 5MW SST 原生 800V 直流试点与无中央冷水机组的冷却设计。
可信度: 66/100
事实核查:
- 无法核实: A wind-powered 800MW campus, 120MW built within a year（原文给出园区规模与建设进度，属机构自述调研，文本内无可核验附件。）
- 无法核实: a 5MW SST-native 800V DC pilot and a cooling design without central chillers（原文描述试点功率、电压与冷却方案，细节具体，需外部资料佐证。）
原文内容:
We Visited China’s Stargate in Inner Mongolia A wind-powered 800MW campus, 120MW built within a year, a 5MW SST-native 800V DC pilot and a cooling design without central chillers https://t.co/t1CnGzAVgG
⏰ 20:18 | ❤️ 374 点赞 | 📝 31词 | 入选: 高赞 374 | 查看原文 →
AI 判断
围绕 AI 能力边界形成两派：核查与研究效率被看好，方法泛化仍被强调为关键。
Ethan Mollick
｜ @emollick
｜ 影响力: 41.5万粉丝 · 互动分 1784.0
Ethan Mollick
｜ @emollick
｜ 影响力: 41.5万粉丝 · 互动分 1784.0
💡 核心观点: 作者认为日常工作中对 AI 输出事实核查的担忧过度，近年模型数据显示 AI 在查找事实和核查断言上已超过人类专家，包括博士级专家。
可信度: 45/100
事实核查:
- 无法核实: data on more recent models suggest they beat human experts, including PhD specialists, in finding facts and checking claims（原文称有数据支持并附链接，但未给出数据集名称、指标或对比结果。）
- 无法核实: It used to be true（作者承认此前担忧成立，但未提供历史依据。）
原文内容:
I think there is way too much concern expressed about verification of AI output in everyday work due to a risk of factual errors. It used to be true, but data on more recent models suggest they beat human experts, including PhD specialists, in finding facts and checking claims. https://t.co/OVlzW3S17n
⏰ 18:50 | ❤️ 1043 点赞 | 📝 52词 | 入选: 高赞 1043 | 查看原文 →
Ethan Mollick
｜ @emollick
｜ 影响力: 41.5万粉丝 · 互动分 1491.77
Ethan Mollick
｜ @emollick
｜ 影响力: 41.5万粉丝 · 互动分 1491.77
💡 核心观点: 作者认为 LLM 能把调查性研究效率提升十倍，分析公司或政府档案这类过去需团队耗费大量时间的工作将迎来调查报道的黄金时代。
可信度: 40/100
事实核查:
- 无法核实: Now LLMs can 10x investigative research（十倍为作者估计，原文未给出测算方法或案例数据。）
- 无法核实: Analyzing filings, records and public data to go deep into a company or government or whatever used to be something that took a team of journalists a long time, so was relatively rare（属对行业现状的描述，原文未附来源。）
原文内容:
This can be a golden age of investigative journalism. Analyzing filings, records and public data to go deep into a company or government or whatever used to be something that took a team of journalists a long time, so was relatively rare. Now LLMs can 10x investigative research.
⏰ 15:03 | ❤️ 912 点赞 | 📝 47词 | 入选: 高赞 912 | 查看原文 →
Bindu Reddy
｜ @bindureddy
｜ 影响力: 39.4万粉丝 · 互动分 717.41
Bindu Reddy
｜ @bindureddy
｜ 影响力: 39.4万粉丝 · 互动分 717.41
💡 核心观点: 作者列举其认定的事实来反驳 AI 相关错误信息：GPU 需求仍高、token 用量随成本下降而爆发、企业未触及 AI 峰值、99% 工作流待自动化，并称 AI 正确率高于几乎所有人、助人提升创造力 10-100 倍、全球层面拉平竞争。
可信度: 34/100
事实核查:
- 无法核实: 99% of workflows still need to be automated（原文给出该比例，但未说明统计口径、样本或来源。）
- 无法核实: AI helps most humans become 10-100x more creative（原文给出倍数区间，未提供测量方法与出处。）
- 无法核实: No, enterprises have NOT hit peak AI（原文为断言式否定，未附企业调研或数据。）
原文内容:
🚨 Tons of misinformation are being spread about AI right now Here are the actual facts - GPU demand is still sky-high - token usage is exploding as costs go down - No, enterprises have NOT hit peak AI - 99% of workflows still need to be automated - AI is more correct than almost all humans - AI helps most humans become 10-100x more creative - Globally, AI has elevated humans and made the playing field more even Overall, this leads to a far more efficient, revenue-maximizing system. It will eventually make work optional and usher in the age of abundance
⏰ 19:07 | ❤️ 462 点赞 | 📝 94词 | 入选: 高赞 462 | 查看原文 →
François Chollet
｜ @fchollet
｜ 影响力: 75.4万粉丝 · 互动分 666.2
François Chollet
｜ @fchollet
｜ 影响力: 75.4万粉丝 · 互动分 666.2
💡 核心观点: 作者主张数据与算力扩展只是方法泛化能力的乘数：巨额投资能让低效、低泛化的方法显得够用，但这不使其成为 AI 的终极答案；应研究可扩展的方法，同时记住更聪明的方法扩展得更好。
可信度: 58/100
事实核查:
- 原文支持: Data scaling and compute scaling are *multipliers* for the generalization power of an approach（原文明确给出该论断，属作者观点，文本中无实验数据。）
- 无法核实: Trillions of dollars in investment can make very inefficient, low-generalization approaches look adequate（原文提及万亿美元级投资规模，未给出具体来源。）
原文内容:
Data scaling and compute scaling are *multipliers* for the generalization power of an approach. Trillions of dollars in investment can make very inefficient, low-generalization approaches look adequate, but that doesn't make them the ultimate end-all-be-all of AI. You should work on methods that *can* scale, because if you don't benefit from this multiplier you simply can't compete. But you should keep in mind scaling is only a multiplier: a cleverer approach will scale *better*.
引用 @RichardSSutton: Brute force computation is not enough for intelligence. Clever new methods are needed. The science of AI is far from finished. IMHO
⏰ 22:52 | ❤️ 406 点赞 | 📝 81词 | 入选: 高赞 406 / 引用讨论 | 查看原文 →
Chubby♨️
｜ @kimmonismus
｜ 影响力: 15.7万粉丝 · 互动分 2229.0
Chubby♨️
｜ @kimmonismus
｜ 影响力: 15.7万粉丝 · 互动分 2229.0
💡 核心观点: 据 AI 2027 追踪器基于 METR 报告中 Anthropic 内部模型的 AECI 分数所做的估算，部分模型已超前于 AI 2027 预测曲线。
可信度: 70/100
事实核查:
- 原文支持: AI 2027 追踪器今日更新（原文称更新，无更多可核验细节）
- 无法核实: 部分模型已领先 AI 2027 曲线（依据为追踪器估算，原文未列出具体分数）
原文内容:
The AI 2027 tracker was updated today. And according to the tracker’s estimates, based on AECI scores from METR’s report on Anthropic’s internal models, some of those models are even ahead of the AI 2027 curve. https://t.co/cjBtmcCuqR
⏰ 07:57 | ❤️ 1662 点赞 | 📝 41词 | 入选: 高赞 1662 | 查看原文 →
商业与争议
预测市场遭执法与资本同时加注，AI 公司增速受质疑，开源项目新增限制引发反弹。
Aakash Gupta
｜ @aakashgupta
｜ 影响力: 33.7万粉丝 · 互动分 9811.5
Aakash Gupta
｜ @aakashgupta
｜ 影响力: 33.7万粉丝 · 互动分 9811.5
💡 核心观点: 犹他州派卧底探员在 Kalshi 下注约 115 美元，Kalshi 两位创始人因此被控 13 项重罪，最高一项可判 15 年；指控公布次日投资者向 Kalshi 注资 10 亿美元。
可信度: 55/100
事实核查:
- 无法核实: 犹他州对 Kalshi CEO 和 COO 提出 13 项非法赌博指控（原文叙述具体，但未附官方文件或链接）
- 无法核实: 指控下达次日投资者向 Kalshi 投资 10 亿美元（原文未给出投资方或报道来源）
- 无法核实: DraftKings 在指控当晚撤出犹他州的体育与政治市场（原文给出时间点，但无外部佐证）
原文内容:
Utah sent an undercover agent to place about $115 in sports bets on Kalshi. Now the two people who started Kalshi are charged with 13 felonies, and the top charge alone carries up to 15 years in prison. The day after the charges came down, investors gave Kalshi $1 billion. Here's how the sting worked. In September, a special agent with the Utah attorney general's office opened a Kalshi account with a Utah address, a Utah phone number and a Utah debit card. Before his first bet, $10 on the Cowboys against the Giants, the app asked him to confirm his location. He told it he was in Utah, and it took his money. He did that 12 times over 11 days, on games like Utah vs. Utah State and a Real Salt Lake match. Kalshi's cut on each bet was somewhere between 3 and 78 cents. So the company made less than $10 off the bets Utah is now using to go after its CEO and COO personally. And that top charge comes from Utah's racketeering law, the same kind of statute prosecutors use on organized crime. The obvious question is why Kalshi didn't just shut off Utah. It clearly knows how. It already blocks Nevada and Washington because a settlement and a court order made it. DraftKings pulled its sports and politics markets out of Utah the same night the charges were filed, FanDuel did it the next day, and Fanatics shut down everything there on November 4. Kalshi stayed because of how the company is built. DraftKings runs on licenses from gambling regulators in more than 20 states, so when one state comes after you with criminal charges, you get out fast and protect the rest. Kalshi has exactly one license, from a federal agency called the CFTC. Its whole $40 billion valuation depends on the idea that this one federal license overrides state gambling laws, and quietly walking away from Utah would cut against that argument everywhere else. The federal government is backing Kalshi on this, and pretty aggressively. The CFTC and the Justice Department sued Utah two days before the charges. Their complaint says that if Kalshi turned away Utah residents, it would be breaking a federal "impartial access" rule that requires exchanges to serve everyone. So the founders are stuck. Utah wants them to stop taking bets from Utahns or face prison, and the federal government is in court saying they aren't allowed to stop. This goes way past Utah, too. The CFTC has sued 10 states since April. On the other side, 37 states signed a brief backing Utah in its appeal. Sequoia and Wellington, who co-led the $1 billion round, are betting on one legal question, whether federal commodities law beats state gambling law. If the courts say yes, Utah's criminal case could fall apart along with every state lawsuit against Kalshi. A federal judge in Utah already said no back in August. The 10th Circuit is up next, and this looks like it's headed to the Supreme Court. Kind of fitting for a company that lets you bet on almost anything that its biggest investors now have $1 billion riding on how a court case turns out.
引用 @DegenTracker_: BREAKING: Utah criminally charges Kalshi's CEO and COO with 13 counts of illegal gambling
⏰ 06:54 | ❤️ 8061 点赞 | 📝 527词 | 入选: 高赞 8061 / 引用讨论 / 长文 | 查看原文 →
Gary Marcus
｜ @GaryMarcus
｜ 影响力: 25.6万粉丝 · 互动分 1092.14
Gary Marcus
｜ @GaryMarcus
｜ 影响力: 25.6万粉丝 · 互动分 1092.14
💡 核心观点: 作者引用 Anthropic 10 月 ARR 超 800 亿美元的报道，称图中绿线是市场幻想、是其维持可持续业务所需的增速，而右侧深色柱显示其已掉队，故判断其可能难以持续。
可信度: 35/100
事实核查:
- 无法核实: Anthropic’s October ARR topped $80B, per WSJ（引自所附推文并称来自 WSJ，原推未附报道链接。）
- 无法核实: The dark bars on the right show that they already fell off the pace（论断依赖未展示的图表，原文无法核对柱状数据。）
- 无法核实: which is why they probably won’t make it（属预测性判断，原文未给出可检验依据。）
原文内容:
Breaking: Green line represents a widespread fantasy, and roughly the rocket that Anthropic would have needed to make it as a sustainable business. The dark bars on the right show that they already fell off the pace — which is why they probably won’t make it. https://t.co/zOOyawVQBX
引用 @tickerplus: Anthropic’s October ARR topped $80B, per WSJ, consistent with TickerTrends tracking
We tracked $82.33B as of Oct. 30, up 7.2% from our Sept. 24 weekly release https://t.co/mz5KCtMbGs
⏰ 14:58 | ❤️ 662 点赞 | 📝 50词 | 入选: 高赞 662 / 引用讨论 | 查看原文 →
Peter Steinberger 🦞
｜ @steipete
｜ 影响力: 61.5万粉丝 · 互动分 1052.99
Peter Steinberger 🦞
｜ @steipete
｜ 影响力: 61.5万粉丝 · 互动分 1052.99
💡 核心观点: 作者认为 Asahi 项目本意是把封闭受限的东西开放，新增限制会妨碍其他项目在其上构建，并称其中的 AI 拒绝机制很荒唐。
可信度: 40/100
事实核查:
- 无法核实: Adding new restrictions that make it harder for other projects to build on that is not the way（属作者对项目政策的观点，原文未列出具体限制条款。）
- 无法核实: And that AI rejection is just silly（原文未说明所指 AI 拒绝的具体场景或证据。）
原文内容:
The Asahi project was all about taking something closed and restricted and opening it up. Adding new restrictions that make it harder for other projects to build on that is not the way. And that AI rejection is just silly. https://t.co/fT4O4wieQm
⏰ 03:02 | ❤️ 799 点赞 | 📝 43词 | 入选: 高赞 799 | 查看原文 →
@levelsio
｜ @levelsio
｜ 影响力: 99.0万粉丝 · 互动分 819.24
@levelsio
｜ @levelsio
｜ 影响力: 99.0万粉丝 · 互动分 819.24
💡 核心观点: 作者引用「最聪明的人周末做的事，就是十年后普通人在工作日做的事」，以此回应科技圈偏好不代表普通用户的质疑。
可信度: 20/100
事实核查:
- 无法核实: What the smartest people (or nerds) do on the weekend is what everyone else will do during the week in ten years（为引用的经验性断言，原文未提供来源或数据。）
- 无法核实: I think you might be overestimating how representative the tech bubble is of ordinary people's preferences（所引推文中的质疑，属观点表达。）
原文内容:
"What the smartest people (or nerds) do on the weekend is what everyone else will do during the week in ten years."
引用 @jonatan_salas: I think you might be overestimating how representative the tech bubble is of ordinary people's preferences. It might be worth talking to people outside those circles.
Just because AI makes something easier to do doesn't necessarily mean people will want to do it themselves. Convenience, time, and …
⏰ 00:21 | ❤️ 593 点赞 | 📝 22词 | 入选: 高赞 593 / 引用讨论 | 查看原文 →
产业与生活
半导体史、3D 重建、开源硬件与育儿感悟并置，呈现技术背后的长期积累与日常视角。
Aakash Gupta
｜ @aakashgupta
｜ 影响力: 33.7万粉丝 · 互动分 2773.0
Aakash Gupta
｜ @aakashgupta
｜ 影响力: 33.7万粉丝 · 互动分 2773.0
💡 核心观点: 讲述手机产业的源头：1916 年波兰化学家 Jan Czochralski 误把笔蘸入熔锡，发现拉速控制单晶生长；贝尔实验室的 Gordon Teal 据此在 1950 年拉出锗单晶。
可信度: 70/100
事实核查:
- 原文支持: Czochralski 于 1918 年发表该发现，用作测量金属凝固速度的方法（原文给出年份与用途，无引用来源）
- 原文支持: 贝尔实验室 1947 年发明晶体管，1950 年 Teal 拉出锗单晶（原文时间线完整，未附文献）
- 无法核实: 1954 年 5 月 Teal 在会议上称口袋里有硅晶体管（原文叙述具体，无外部佐证）
原文内容:
Your phone exists because a tired guy dipped his pen in the wrong cup back in 1916. He died making shoe polish, and they buried him in a grave without his name on it. His name was Jan Czochralski, a Polish chemist working late one night at a lab in Berlin. He was writing up his notes, reached for his inkwell, and dunked his pen into a pot of melted tin instead. When he pulled it out, a thin thread of metal was hanging off the tip. He forgot about his notes and spent the rest of the night dipping the pen back in, pulling faster, pulling slower, until he figured out the speed controlled how long the thread got. When he looked closer, the whole thread was one single crystal. He wrote it up in 1918 as a way to measure how fast metals harden, and then it sat in a German chemistry journal for about 30 years. Then Bell Labs invented the transistor in 1947 and ran straight into a problem. The ones they built out of clumpy, multi-grain material were all over the place. To work reliably, transistors needed one perfect crystal, with every atom lined up in the same pattern. A Bell Labs scientist named Gordon Teal was convinced the pen trick was the answer. His bosses weren't interested, so he worked on it after hours with barely any resources. He'd lower a tiny seed crystal into molten germanium, spin it, and pull it up really slowly. By 1950 he had it working. Then he left for Texas Instruments, and in May 1954 he showed up at a conference where the experts were saying silicon transistors were years off and told everyone he had a few in his pocket. Getting sand to 98% pure, like in this video, is honestly the easy part. A chip needs silicon at least 99.9999999% pure, and the best stuff gets to eleven nines. That works out to one wrong atom for every 100 billion. Then it's melted at 1,414°C and pulled up out of a quartz bowl as one flawless crystal about a foot wide, using basically the same motion Czochralski made with his pen. Even that bowl comes with a catch. Pretty much all the quartz pure enough to hold molten silicon comes from two mines in Spruce Pine, North Carolina, a town of about 2,000 people. Hurricane Helene knocked both of them offline in 2024. Czochralski never got to see any of this. When World War II ended he was arrested and accused of collaborating with the Nazis, who had taken over his lab. Prosecutors dropped the case that August, and his university kicked him out anyway. He moved back to his small hometown and ran a little company making shoe polish and cosmetics until a heart attack killed him in 1953. It took until 2011 for archive documents to surface showing he'd been quietly helping the Polish resistance during the war, putting fighters on his payroll and making parts for their weapons. His old university cleared his name that year, 58 years after he died. Every advanced chip being made today, including the ones running AI, still gets grown with the trick he found by accident that night.
引用 @thekitze: humans are MENTAL bro https://t.co/uD4C1S3Boz
⏰ 08:43 | ❤️ 2035 点赞 | 📝 532词 | 入选: 高赞 2035 / 引用讨论 / 长文 | 查看原文 →
Meng To
｜ @MengTo
｜ 影响力: 19.0万粉丝 · 互动分 1068.13
Meng To
｜ @MengTo
｜ 影响力: 19.0万粉丝 · 互动分 1068.13
💡 核心观点: 作者称让 Opus 5.5 用 three.js 重建自己的房间，用于制作教程动画，并感叹 3D 与动态设计能力进步之大。
可信度: 40/100
事实核查:
- 无法核实: I asked Opus 5.5 to recreate my room in three.js just so it can animate my tutorial videos（为作者个人使用经历，原文附有链接但未展示成品细节。）
- 无法核实: 3D + motion design. Insane how far we've come.（属主观评价，无可核验数据。）
原文内容:
I asked Opus 5.5 to recreate my room in three.js just so it can animate my tutorial videos. 3D + motion design. Insane how far we've come. https://t.co/KJl8yAVANt
⏰ 16:22 | ❤️ 832 点赞 | 📝 29词 | 入选: 高赞 832 | 查看原文 →
swyx 🔜 aie.codes
｜ @swyx
｜ 影响力: 21.2万粉丝 · 互动分 676.73
swyx 🔜 aie.codes
｜ @swyx
｜ 影响力: 21.2万粉丝 · 互动分 676.73
💡 核心观点: 作者称收到测试订单后惊喜发现该客户端开源，认为这对插入物理端口的设备很重要；并表示终于用上闲置 SD 卡槽，API 设计良好且对 agent 友好，预告将在 aiDotEngineer 有惊喜。
可信度: 55/100
事实核查:
- 无法核实: this client is open source（原文直接陈述，未给出仓库地址或开源协议等可核验信息。）
- 原文支持: the api is so nice and agent friendly（原文为主观评价，无具体接口细节支撑。）
原文内容:
just got my test order and was delightfully surprised to find that this client is open source!!!! (v impt for sticking anything in a physical port) very delighted to have a use for this useless SD Card slot and the api is so nice and agent friendly. watch for surprises at @aiDotEngineer ...
引用 @pkuhar: Do you have a MacBook Pro? With an SD Card slot?
Your slot now has an API. Write to a file, blink the lights.
Or let your agent use it to get your attention.
More into https://t.co/VYFXVO4rR2
LED Animation Format: https://t.co/DXPRb4qtSB https://t.co/3uYVeeYdVY
⏰ 09:18 | ❤️ 372 点赞 | 📝 52词 | 入选: 高赞 372 / 引用讨论 | 查看原文 →
GREG ISENBERG
｜ @gregisenberg
｜ 影响力: 75.9万粉丝 · 互动分 682.47
GREG ISENBERG
｜ @gregisenberg
｜ 影响力: 75.9万粉丝 · 互动分 682.47
💡 核心观点: 作者分享育儿体验：透过孩子重新经历童年，从怕黑、等生日到把纸箱当最好的玩具，如今站在另一侧为这些时刻营造安全与奇妙，并以初次看世界的视角重拾好奇。
可信度: 20/100
事实核查:
- 原文支持: I didn't expect parenting to make me feel younger on one hand, and older on the other（原文为个人感受陈述，属主观体验，无外部可核验内容。）
原文内容:
I think the coolest part of having kids is reliving your own childhood in a different way. You remember what it felt like to be scared of the dark, to wait all year for a birthday or to think a cardboard box was the best toy in the world. The difference is that now you're on the other side of it. And you get to be the person who makes those moments feel safe and magical. You see the world through someone who's seeing it for the first time, and it makes you curious again. I didn't expect parenting to make me feel younger on one hand, and older on the other but some days it does. Highly recommend :)
⏰ 23:41 | ❤️ 502 点赞 | 📝 122词 | 入选: 高赞 502 / 长文 | 查看原文 →
🔗 引用关系
- @aakashgupta 引用 @DegenTracker_：犹他州派卧底探员在 Kalshi 下注约 115 美元，Kalshi 两位创始人因此被控 13 项重罪，最高一项可判 15 年；指控公布次日投资者向 Kalshi 注资 10 亿美元。（原文：BREAKING: Utah criminally charges Kalshi's CEO and COO with 13 counts of illegal gambling）
- @karpathy 引用 @karpathy：引用 Tristan 的 YouTube 频道榜单，用 Claude Code 三分钟做出 TristanTube，汇集 4.7 万条硬核视频并提供电视模式。（原文：Weekend hack: HackerTube. I was annoyed with YouTube's recommendations so I had Claude filter down 43M comments over 20 …）
- @aakashgupta 引用 @thekitze：讲述手机产业的源头：1916 年波兰化学家 Jan Czochralski 误把笔蘸入熔锡，发现拉速控制单晶生长；贝尔实验室的 Gordon Teal 据此在 1950 年拉出锗单晶。（原文：humans are MENTAL bro https://t.co/uD4C1S3Boz）
- @gregisenberg 引用 @RoundtableSpace：作者引用「Apple 把 Xcode 放进 Claude Code」的消息，给出用 AI 代理选品、验证需求、开发与测试 iOS 应用的清单。（原文：Apple just put Xcode inside Claude Code.
It's an official Apple plugin, now in Claude Code's plugin directory:
• Bu…）
- @GaryMarcus 引用 @tickerplus：作者引用 Anthropic 10 月 ARR 超 800 亿美元的报道，称图中绿线是市场幻想、是其维持可持续业务所需的增速，而右侧深色柱显示其已掉队，故判断其可能难以持续。（原文：Anthropic’s October ARR topped $80B, per WSJ, consistent with TickerTrends tracking
We tracked $82.33B as of Oct. 30, u…）
- @levelsio 引用 @jonatan_salas：作者引用「最聪明的人周末做的事，就是十年后普通人在工作日做的事」，以此回应科技圈偏好不代表普通用户的质疑。（原文：I think you might be overestimating how representative the tech bubble is of ordinary people's preferences. It might be …）
- @marclou 引用 @marclou：作者称“黑色十月”后采取更多措施：移除几乎全部浏览器扩展、把仓库依赖精简到最低限度、设置 Grok 机器人每小时扫描网络，发现安全问题就提交 PR 并发送 Telegram 提醒。（原文：I spent the last 48 hours rotating passwords, adding 2FA everywhere, and spawning agents to audit all my startups' codeb…）
- @swyx 引用 @pkuhar：作者称收到测试订单后惊喜发现该客户端开源，认为这对插入物理端口的设备很重要；并表示终于用上闲置 SD 卡槽，API 设计良好且对 agent 友好，预告将在 aiDotEngineer 有惊喜。（原文：Do you have a MacBook Pro? With an SD Card slot?
Your slot now has an API. Write to a file, blink the lights.
Or let y…）
- @fchollet 引用 @RichardSSutton：作者主张数据与算力扩展只是方法泛化能力的乘数：巨额投资能让低效、低泛化的方法显得够用，但这不使其成为 AI 的终极答案；应研究可扩展的方法，同时记住更聪明的方法扩展得更好。（原文：Brute force computation is not enough for intelligence. Clever new methods are needed. The science of AI is far from fin…）
🧩 跨推文归纳
共识
- 多条推文印证 Claude Code 已成为通用代理入口：建站、剪辑工具、家具设计、iOS 开发与维权均在其内完成。
- 开源与离线可运行被视为降低门槛的关键，个人设备与免费工具被反复提及。
分歧
- Ethan Mollick 认为 AI 事实核查已超人类专家，François Chollet 则强调数据与算力只是低泛化方法的乘数。
- Gary Marcus 质疑 Anthropic 高 ARR 的可持续性，Bindu Reddy 则列举需求与用量仍高的判断予以反驳。
数据来源：X 列表「AI 英文」（22 条候选）· 由 twscrape 采集、grok 整理
© 版权声明
文章版权归作者所有，未经允许请勿转载。
相关文章
暂无评论...