共收录 22 篇深度内容
📌 今日概览
- 算力供给成为焦点：数据中心收紧传闻下，SpaceX 对 Anthropic 的算力协议被指可提前终止并涨价续约。
- 芯片侧给出具体对比：AMD MI355 借 TileRT 软件在长上下文负载下快于 GB300 NVL72。
- 模型评测引入置信度校准新指标，决策模型头部差距缩至 1 分以内。
- 就业与泡沫争论持续，CEO 的宏观预测被指与生产率、就业数据不符。
🔥 必读 Top3
- Artificial Analysis 作者介绍 Artificial Analysis Decision Index v2：采用新指标同时奖励正确性与置信度校准，私有数据集占比升至 45%。据其介绍，Drex Max v1.1 以 73 分在决策模型中领先，与 GLiDE v1.2、Mammoth 27B 相差不到 1 分。
入选理由：以新指标同时奖励正确性与置信度校准，私有数据集占比升至 45%，给出决策模型的新排序。 - Rohan Paul 作者转述Chamath观点：若数据中心供给收紧，SpaceX可提前90天终止与Anthropic的算力协议并以2至3倍价格续约，Anthropic只能接受。
入选理由：转述供给收紧下 SpaceX 可提前 90 天终止与 Anthropic 算力协议并涨价续约，揭示算力交易的议价结构。 - SemiAnalysis 作者称 AMD MI355 借助 TileRT 软件在长上下文负载下比 GB300 NVL72 快1.42倍，达到每用户 p90 1330 tok/s，并感谢 AMD 提供 MI350 集群。
入选理由：实测 AMD MI355 借 TileRT 在长上下文比 GB300 NVL72 快 1.42 倍，提供可核对的性能数字。
🧭 主题分组
- 算力博弈（2 条）—— 供给收紧预期下，算力协议与芯片性能对比同时成为议价筹码。
- 模型评测（1 条）—— 决策能力评测转向奖励校准，头部模型分数高度接近。
- 就业争论（2 条）—— AI 是否导致失业分歧明显，一方引衰退与高失业率，一方指生产率与就业几乎未变。
- 市场情绪（2 条）—— 科技股回调与泡沫风险被反复讨论，IPO 时机与成本下降预期并存。
- 模型走向（3 条）—— 一边押注小模型普及，一边认为现有算力足以走向超级智能。
- 工具生态（3 条）—— 智能体与编程工具的实用价值成为主线，移动端接入被重点提及。
- 创意提示（2 条）—— 提示词与生成工具继续产出可视化玩法。
- 人事传闻（1 条）—— 一则低可信度人事传闻与市场波动被并列讨论。
- 生活话题（5 条）—— 非技术内容集中在幸福感来源与日常趣味见闻。
- 其他（1 条）—— 未归类的补充内容
📖 详细内容
算力博弈
供给收紧预期下，算力协议与芯片性能对比同时成为议价筹码。
Rohan Paul
｜ @rohanpaul_ai
｜ 影响力: 16.1万粉丝 · 互动分 1284.0
Rohan Paul
｜ @rohanpaul_ai
｜ 影响力: 16.1万粉丝 · 互动分 1284.0
💡 核心观点: 作者转述Chamath观点：若数据中心供给收紧，SpaceX可提前90天终止与Anthropic的算力协议并以2至3倍价格续约，Anthropic只能接受。
可信度: 40/100
事实核查:
- 原文支持: SpaceX could end its compute deal with Anthropic on 90 days notice and bring it back at 2x to 3x the price（原文明确转述该说法，属引用他人观点。）
- 无法核实: Anthropic would have to take it（原文给出该推断，但无合同条款或数据支撑。）
原文内容:
Chamath says that if data center supply tightens, SpaceX could end its compute deal with Anthropic on 90 days notice and bring it back at 2x to 3x the price, and Anthropic would have to take it. "So, whatever revenue you see on SpaceX's balance sheet from Ant could two to 3x and ANT would have no choice except to immediately capitulate and take the higher price. It would also show all of the other hyperscalers that they are very far away from having economic yield that is optimized. And so then it forces them to consider should I be giving it to myself? Should I be selling it to the other people? It then forces the Neoclouds to take all of this money that they've raised, whether it's debt or equity, and say, "I have two choices, which is to buy as much compute today as I can at whatever price it's available because the other path is assured destruction and bankruptcy for me because I am just out of the game." And so, what do they do? They will take the last gasp and do whatever they have. So all the game theory comes down to this one deal and whether Elon decides to rip it because if he rips it, watch out. ---- Full video on "All-In Podcast" YouTube channel, (link in comment)
⏰ 12:41 | ❤️ 922 点赞 | 📝 226词 | 入选: 高赞 922 / 长文 | 查看原文 →
SemiAnalysis
｜ @SemiAnalysis_
｜ 影响力: 18.4万粉丝 · 互动分 588.64
SemiAnalysis
｜ @SemiAnalysis_
｜ 影响力: 18.4万粉丝 · 互动分 588.64
💡 核心观点: 作者称 AMD MI355 借助 TileRT 软件在长上下文负载下比 GB300 NVL72 快1.42倍，达到每用户 p90 1330 tok/s，并感谢 AMD 提供 MI350 集群。
可信度: 60/100
事实核查:
- 无法核实: AMD MI355 is 1.42x FASTER THAN GB300 NVL72（原文给出倍数与对比对象，未附测试方法、配置或原始报告。）
- 无法核实: 1330 p90 tok/s/user on long context workloads（原文给出性能指标，未说明测试环境与数据集，暂无法独立核验。）
原文内容:
ALERT🚨🚨: AMD MI355 is 1.42x FASTER THAN GB300 NVL72. THIS IS ACHIEVED USING THE TILERT SOFTWARE, WHICH GETS 1330 p90 tok/s/user on long context workloads! Thanks to AMD leadership for providing an MI350 cluster to TileRT maintainers, enabling them to achieve this great result!
⏰ 19:50 | ❤️ 407 点赞 | 📝 39词 | 入选: 高赞 407 | 查看原文 →
模型评测
决策能力评测转向奖励校准，头部模型分数高度接近。
Artificial Analysis
｜ @ArtificialAnlys
｜ 影响力: 17.6万粉丝 · 互动分 563.92
Artificial Analysis
｜ @ArtificialAnlys
｜ 影响力: 17.6万粉丝 · 互动分 563.92
💡 核心观点: 作者介绍 Artificial Analysis Decision Index v2：采用新指标同时奖励正确性与置信度校准，私有数据集占比升至 45%。据其介绍，Drex Max v1.1 以 73 分在决策模型中领先，与 GLiDE v1.2、Mammoth 27B 相差不到 1 分。
可信度: 72/100
事实核查:
- 原文支持: 45% of the Index is now built on private datasets, up from 25% in v1（原文直接给出 v2 私有数据集占比 45%、v1 为 25%，数字与对比均可在原文找到。）
- 无法核实: Drex Max v1.1 from @NaceAI narrowly leads among decision models at 73, within 1 point of GLiDE v1.2 (@fastinoAI) and Mammoth 27B (@baselabs)（原文给出 Drex Max v1.1 的 73 分及与 GLiDE v1.2、Mammoth 27B 相差不到 1 分的排名描述，未提供各家具体分数。）
原文内容:
Introducing the Artificial Analysis Decision Index v2: a new proprietary evaluation, updated benchmarks, and a headline metric that rewards model calibration We’ve released the second version of the Artificial Analysis Decision Index to better reflect real-world utility and guard against benchmark gaming. 45% of the Index is now built on private datasets, up from 25% in v1. Drex Max v1.1 from @NaceAI narrowly leads among decision models at 73, within 1 point of GLiDE v1.2 (@fastinoAI) and Mammoth 27B (@baselabs) Updates in Decision Index v2: ➤ Decision Score: We adopt a new headline metric that rewards a model for both correctness and calibrated confidence. Each question scores from 0 to 100: right answers score above 50, wrong answers below 50, and confidence in its top answer moves the score within each half ➤ AA-Paperclip (30%): We added a new private benchmark of 1,500 documents including resumes, receipts, invoices, and AI and machine learning research papers. Models make decisions and extract facts from each document and are scored jointly as a single decision, correct only if all 15 answers are right ➤ AA-Symmetry (15%): This evaluates whether models change their answers for equivalent questions worded differently. We test 120 questions with variations such as reordered options, swapped labels, and negated yes/no questions, then measure how consistent the model’s probabilities are across those variations ➤ AA-Omniscience-Decision v1.1 and HLE-MultiChoice v1.1 (15% each, down from 25%): AA-Omniscience-Decision now varies the wording of its “I don’t know” option across ten alternatives. HLE-MultiChoice adds an option to decline using the same ten phrasings We’ve also added EvalSafe-ONET-AA (15%), our implementation of @typesafeai’s ONET benchmark. We have retained When2Call from @NVIDIAAI in the Index at a decreased weighting of 10% (down from 25%), and removed WildGuardTest given its high saturation Key results: ➤ Three decision models sit within 1 point at the top of the Index: Drex Max v1.1 from Nace AI and GLiDE v1.2 from Fastino Labs both score 73, with Mammoth 27B from Base Labs scoring 72 ➤ Cost per 1M Tasks varies by more than 60x across decision models: Celeris-1.1 Decision ($14), Decider v1.2 ($68) and Drex Max v1.1 ($102) sit on the Decision Index vs. Cost Pareto frontier, while Clef costs $859. Token price is not the only cost driver: most decision models, including Drex Max v1.1 and Decider v1.2, bill a document’s input once for all its questions, while GLiDE v1.2 and Decider Flash v1.2 bill it for every question. This means GLiDE v1.2 costs $780, nearly 8x higher than Drex Max v1.1 despite only 33% higher list prices ➤ Language models can score higher, alongside higher Cost and Time per Task: Claude Haiku 5.5 (Max) scores 77 and GPT-6 Luna (Max) scores 76 on Decision Index v2, putting them ~3 points ahead of Drex Max v1.1. GPT-6 Luna (Max) costs $1,530 per 1M Tasks vs. $102 for Drex Max v1.1 (15x the cost) and takes 11.34 seconds per task vs. 275 milliseconds (~40x slower)
⏰ 23:51 | ❤️ 316 点赞 | 📝 441词 | 入选: 高赞 316 / 长文 | 查看原文 →
就业争论
AI 是否导致失业分歧明显，一方引衰退与高失业率，一方指生产率与就业几乎未变。
Gary Marcus
｜ @GaryMarcus
｜ 影响力: 25.7万粉丝 · 互动分 1723.48
Gary Marcus
｜ @GaryMarcus
｜ 影响力: 25.7万粉丝 · 互动分 1723.48
💡 核心观点: 作者引用Anthropic CEO关于经济年增10%、20%人口失业的预测，指出就业与生产率几乎未变，批评CEO随口给出数字。
可信度: 46/100
事实核查:
- 无法核实: jobs have barely moved; productivity has barely changed（原文给出该判断，但未附数据来源。）
- 原文支持: the economy grows 10% a year and 20% of people don't have jobs（原文明确转述该预测，引用关系清楚。）
原文内容:
jobs have barely moved; productivity has barely changed. and yet we just listen to CEOs who pull numbers out of a hat.
引用 @0xFazerAI: Anthropic's CEO described a future where the economy grows 10% a year and 20% of people don't have jobs.
In the video below, Anderson Cooper reads it back to him. His answer: "That's exactly what I was getting to."
He says AI could wipe out half of entry-level white-collar jobs within one to five …
⏰ 12:33 | ❤️ 1318 点赞 | 📝 22词 | 入选: 高赞 1318 / 引用讨论 | 查看原文 →
Gary Marcus
｜ @GaryMarcus
｜ 影响力: 25.7万粉丝 · 互动分 736.7
Gary Marcus
｜ @GaryMarcus
｜ 影响力: 25.7万粉丝 · 互动分 736.7
💡 核心观点: 作者引用 Paul Krugman 观点称，AI 泡沫破裂几乎必然引发衰退，且失业率将多年居高不下，以此论证 AI 会导致失业。
可信度: 40/100
事实核查:
- 无法核实: a burst AI bubble will almost surely cause a recession（该判断为被引用者的预测性观点，原文未提供模型或数据支撑。）
- 无法核实: unemployment will remain elevated for multiple years（原文引述为对未来走势的推断，缺少可核验的历史对照数据。）
原文内容:
The other reason AI is going to lead to unemployment, today from @paulkrugman: “a burst AI bubble will almost surely cause a recession. The depth of that recession is highly uncertain, although I favor the view that it will be worse than the slump at the end of the 1990s tech boom. But even if the recession is short and shallow, history and theory suggest that recovery from that slump will be painfully slow, that unemployment will remain elevated for multiple years.”
⏰ 16:20 | ❤️ 444 点赞 | 📝 81词 | 入选: 高赞 444 | 查看原文 →
市场情绪
科技股回调与泡沫风险被反复讨论，IPO 时机与成本下降预期并存。
Nathan Lambert
｜ @natolambert
｜ 影响力: 10.9万粉丝 · 互动分 459.9
Nathan Lambert
｜ @natolambert
｜ 影响力: 10.9万粉丝 · 互动分 459.9
💡 核心观点: 作者判断科技股若因 Jensen 相关消息继续急跌，本轮行情可能迎来首次大幅回调，并提及债务过热、选举与 Jensen 因素接连出现，以及 Anthropic 或在低位 IPO。
可信度: 28/100
事实核查:
- 无法核实: Debt went too far, the election, Jensen all in short succession… and Anthropic gonna IPO at the bottom?（原文仅作并列归因与疑问式推测，未给出债务规模、选举或 Jensen 事件的具体信息。）
原文内容:
If tech stocks continue sharply downward on the Jensen news, looking back this could be what we think of as the first major correction in this boom. Debt went too far, the election, Jensen all in short succession… and Anthropic gonna IPO at the bottom?
⏰ 21:04 | ❤️ 367 点赞 | 📝 45词 | 入选: 高赞 367 | 查看原文 →
@levelsio
｜ @levelsio
｜ 影响力: 99.1万粉丝 · 互动分 568.03
@levelsio
｜ @levelsio
｜ 影响力: 99.1万粉丝 · 互动分 568.03
💡 核心观点: 作者回应称 AI 正在降低一切成本，先节省时间再降低周边开销，并重申会让所有东西便宜5到20倍。
可信度: 22/100
事实核查:
- 无法核实: I keep saying it'll make everything 5x to 20x cheaper（原文为个人预测性断言，未提供测算依据或案例。）
原文内容:
Same, it's lowering the cost of everything, first in time (we have more than time than ever) and then in cost of everything around us I keep saying it'll make everything 5x to 20x cheaper
引用 @adamdotdev: I’m at least 100x more productive than I was before AI. And that’s with a much better quality of life; I can get more done in a day running errands than I did in a month of being chained to my desk.
If AI doesn’t kill us all, I can’t help but think we’re headed for some really great times.
⏰ 19:53 | ❤️ 386 点赞 | 📝 35词 | 入选: 高赞 386 / 引用讨论 | 查看原文 →
模型走向
一边押注小模型普及，一边认为现有算力足以走向超级智能。
Bindu Reddy
｜ @bindureddy
｜ 影响力: 39.4万粉丝 · 互动分 432.97
Bindu Reddy
｜ @bindureddy
｜ 影响力: 39.4万粉丝 · 互动分 432.97
💡 核心观点: 作者预测：算力稀缺且昂贵，大模型难以无处不在，小型模型会快速进步并用于一切场景，从而主导 AI。
可信度: 30/100
事实核查:
- 无法核实: Small models will rapidly improve and be used for everything（原文为作者的未来预测，无数据或来源支撑。）
原文内容:
PREDICTION - Small models will dominate AI With compute being super scarce and expensive It will become impossible to use big models ubiquitously Small models will rapidly improve and be used for everything 🔥
⏰ 16:20 | ❤️ 220 点赞 | 📝 32词 | 入选: 高赞 220 | 查看原文 →
Rohan Paul
｜ @rohanpaul_ai
｜ 影响力: 16.1万粉丝 · 互动分 363.48
Rohan Paul
｜ @rohanpaul_ai
｜ 影响力: 16.1万粉丝 · 互动分 363.48
💡 核心观点: 作者引用并转述 Alex Wissner-Gross 的观点：AI 已达「超级智能逃逸速度」，即便立刻停止所有数据中心建设，凭借现有算力与算法洞察仍会走向超级智能。
可信度: 55/100
事实核查:
- 无法核实: We're at super intelligence escape velocity at this point and stop all data center construction. Yes, it would be painful, but yes, I I think we have enough algorithmic insight with combined with the compute that's already available.（原文为受访者观点性断言，未给出可核验数据；作者注明了 Peter H. Diamandis 的 YouTube 频道 Moonshot 播客作为出处。）
原文内容:
Physicist, computer scientist Dr. Alex Wissner-Gross (@alexwg): AI has reached “superintelligence escape velocity,” so even halting every new data center tomorrow would not stop the path to superintelligence. "We're at super intelligence escape velocity at this point and stop all data center construction. Yes, it would be painful, but yes, I I think we have enough algorithmic insight with combined with the compute that's already available." ---- Full video on "Peter H. Diamandis" YouTube channel (Moonshot podcast), link in comment
⏰ 15:06 | ❤️ 198 点赞 | 📝 82词 | 入选: 高赞 198 | 查看原文 →
Chubby♨️
｜ @kimmonismus
｜ 影响力: 15.8万粉丝 · 互动分 1823.0
Chubby♨️
｜ @kimmonismus
｜ 影响力: 15.8万粉丝 · 互动分 1823.0
💡 核心观点: 作者称部分用户已被路由到GPT-6.2 Sol，并预测下周将是正式发布周，引用内容展示单提示生成Three.js场景的效果。
可信度: 34/100
事实核查:
- 无法核实: users are being routed to GPT-6.2 Sol（原文称有此现象，但未给出截图或官方说明。）
- 原文支持: These aren't screenshots from a game. They're Three.js scenes generated from a single prompt, with everything packed into ONE HTML file（引用内容描述了生成结果，属作者转述。）
原文内容:
It seems all over, users are being routed to GPT-6.2 Sol. Last week was a disappointment, no new model released. But now it looks like next week is finally the release week. And check out the outputs below. Crazy.
引用 @TokenGremlin: GPT-6.2 Sol is already showing up for some ChatGPT users, and just look at what it's creating.
These aren't screenshots from a game. They're Three.js scenes generated from a single prompt, with everything packed into ONE HTML file.
I've been waiting to see what Sol could do. And now I think we're …
⏰ 12:40 | ❤️ 1480 点赞 | 📝 39词 | 入选: 高赞 1480 / 引用讨论 | 查看原文 →
工具生态
智能体与编程工具的实用价值成为主线，移动端接入被重点提及。
Machina
｜ @EXM7777
｜ 影响力: 15.8万粉丝 · 互动分 1062.0
Machina
｜ @EXM7777
｜ 影响力: 15.8万粉丝 · 互动分 1062.0
💡 核心观点: 作者基于超1000小时使用经验称Hermes是其最终选择的AI智能体，列举其迭代速度、模型兼容与故障修复优势，并看好其移动端。
可信度: 28/100
事实核查:
- 无法核实: i've spent over 1,000 hours building with AI agents（原文给出该数字，无法独立核验。）
- 原文支持: one 'hermes doctor' fixes it（原文明确描述该修复命令。）
- 无法核实: sometimes more efficient than Codex or Claude Code for me（原文为主观比较，未给出测试数据。）
原文内容:
i've spent over 1,000 hours building with AI agents... and i've tried them ALL from OpenClaw to Grok Bot, ChatGPT Dots and Open Bot... every single time, i end up back on Hermes here's why: - it hasn't stopped evolving since i first tried it in march, it keeps absorbing the best features from its competitors - the team behind it is locked in, new models, connectors and plugins land extremely fast - it's one of the best harnesses you can use, sometimes more efficient than Codex or Claude Code for me - you can run any model, bring your own subscriptions, or run it locally and it picks the best model for your machine - it just works... and when it breaks, one 'hermes doctor' fixes it i get why people are impressed by Grok Bot's UI, but it takes 5 minutes to make the Hermes desktop app look exactly how you want hide the tool calls, the extra menus, everything you don't need, and you end up with a setup that's yours forever the mobile app is in beta right now, i should get access in a few hours, and what i've seen so far is legit once it goes public, a lot of people will realize no other personal AI agent can compete so the time you put into your Hermes setup today pays off hard in the coming months because when Grok Bot runs on a nerfed model, when Muse gets locked to the US, or when your usage limits everywhere else get ridiculous... Hermes will still be standing and delivering, like it has for me for the past 6 months
⏰ 14:50 | ❤️ 710 点赞 | 📝 276词 | 入选: 高赞 710 / 长文 | 查看原文 →
Riley Brown
｜ @rileybrown
｜ 影响力: 26.4万粉丝 · 互动分 309.41
Riley Brown
｜ @rileybrown
｜ 影响力: 26.4万粉丝 · 互动分 309.41
💡 核心观点: 作者列举 Claude Duet 连接 Mac 上 Claude Code 的五个用途：虚拟与实体电脑、手机剪辑视频、手机建应用、手机端电脑操作、与 Claude Code 实时语音。
可信度: 55/100
事实核查:
- 无法核实: Claude Duet can now connect to Claude Code running on my Mac（原文陈述该连接能力并列出五条用途与时间戳，属产品功能介绍，未给出可独立核验的细节。）
原文内容:
Claude Duet can now connect to Claude Code running on my Mac. This is useful for these 5 reasons 00:00 Intro 00:22 1. Virtual + Physical Computer 03:24 2. Edit Videos From Your Phone 06:56 3. Build Apps From Your Phone 08:48 4. Computer Use From Your Phone 10:29 5. Real-Time Voice With Claude Code
⏰ 21:04 | ❤️ 201 点赞 | 📝 43词 | 入选: 高赞 201 | 查看原文 →
swyx 🔜 aie.codes
｜ @swyx
｜ 影响力: 21.2万粉丝 · 互动分 574.37
swyx 🔜 aie.codes
｜ @swyx
｜ 影响力: 21.2万粉丝 · 互动分 574.37
💡 核心观点: 作者认为 Cursor 被收购后表现强劲，称其2025年 ARR 达10亿美元、企业业务环比增长300%，并主张以 SEC 审计财报判断现实。
可信度: 58/100
事实核查:
- 无法核实: cursor ended 2025 arr 1B arr（原文给出具体 ARR 数字，但未附财报或官方公告链接。）
- 无法核实: enterprise growing 300% QoQ（原文给出环比增速，缺少数据口径与来源说明。）
- 无法核实: twitter vibes are a funhouse mirror of actual reality that you can now basically look up in SEC audited financials（原文主张可查 SEC 审计财报，但未给出具体文件编号。）
原文内容:
occasional reminder that cursor is slaying it esp post acquisition (prob one of the best that elon has ever done, better than twitter itself) and twitter vibes are a funhouse mirror of actual reality that you can now basically look up in SEC audited financials cursor ended 2025 arr 1B arr btw. enterprise growing 300% QoQ
引用 @skeptrune: @netspencer dead?
⏰ 13:58 | ❤️ 319 点赞 | 📝 53词 | 入选: 高赞 319 / 引用讨论 | 查看原文 →
创意提示
提示词与生成工具继续产出可视化玩法。
Amira Zairi
｜ @azed_ai
｜ 影响力: 6.1万粉丝 · 互动分 620.36
Amira Zairi
｜ @azed_ai
｜ 影响力: 6.1万粉丝 · 互动分 620.36
💡 核心观点: 作者分享一套生成3D名人夸张漫画形象的提示词，要求夸张头部比例、真实皮肤纹理与影棚布光等细节。
可信度: 65/100
事实核查:
- 原文支持: Prompt share: 3D Celebrity Caricature（原文完整给出提示词文本，可直接复制使用，属可核验的操作性内容。）
原文内容:
Prompt share: 3D Celebrity Caricature 💬Prompt: A hyperrealistic 3D caricature of [celebrity], oversized head, exaggerated yet recognizable facial features, authentic skin pores, fine wrinkles, natural blemishes, subsurface scattering, realistic eyes, detailed hair strands, [signature outfit], [expressive pose], humorous proportions, lifelike human skin shaders, high-resolution PBR textures, photorealistic CGI rendering, cinematic studio lighting, full-body character, clean white background Try it and share your art ✨
⏰ 10:57 | ❤️ 447 点赞 | 📝 63词 | 入选: 高赞 447 | 查看原文 →
@levelsio
｜ @levelsio
｜ 影响力: 99.1万粉丝 · 互动分 987.5
@levelsio
｜ @levelsio
｜ 影响力: 99.1万粉丝 · 互动分 987.5
💡 核心观点: 作者推荐独立开发者 @apekshik，称其作品都很酷，最新作品是一个无限演化的 Backrooms 模拟器，并附链接。
可信度: 55/100
事实核查:
- 原文支持: His latest is an infinitely evolving Backrooms Simulator（原文称其为该作者最新作品并给出链接，可在链接页面核验。）
- 无法核实: 35,000 players and ~9,600 hours played（数据来自被引推文作者自述，原文未提供平台后台或第三方统计。）
原文内容:
Every project this guy makes is very cool Very good follow: @apekshik His latest is an infinitely evolving Backrooms Simulator: https://t.co/RM0g6MSefF
引用 @apekshik: i'm a solo dev and i built an infinitely evolving backrooms game and released it on the web.
35,000 players and ~9,600 hours played. these are my metrics under 36 hours of launch.
each blue dot below is a player on the map i recorded a moment ago. there's ~1,200 players playing the game at THIS …
⏰ 20:20 | ❤️ 783 点赞 | 📝 23词 | 入选: 高赞 783 / 引用讨论 | 查看原文 →
人事传闻
一则低可信度人事传闻与市场波动被并列讨论。
Anthony Pompliano 🌪
｜ @APompliano
｜ 影响力: 244.0万粉丝 · 互动分 3584.0
Anthony Pompliano 🌪
｜ @APompliano
｜ 影响力: 244.0万粉丝 · 互动分 3584.0
💡 核心观点: 作者称英伟达CEO黄仁勋承认与一名被美国政府怀疑为中国间谍的女性有婚外情，并以戏剧化语气评论此事。
可信度: 22/100
事实核查:
- 无法核实: Nvidia CEO Jensen Huang has admitted that he had an affair with a woman who the US government suspects is a Chinese spy（原文给出该主张，但未提供任何来源或可核验细节。）
原文内容:
Wow. Nvidia CEO Jensen Huang has admitted that he had an affair with a woman who the US government suspects is a Chinese spy. You couldn't write a Hollywood script better than reality.
⏰ 20:50 | ❤️ 2898 点赞 | 📝 34词 | 入选: 高赞 2898 | 查看原文 →
生活话题
非技术内容集中在幸福感来源与日常趣味见闻。
Anthony Pompliano 🌪
｜ @APompliano
｜ 影响力: 244.0万粉丝 · 互动分 858.38
Anthony Pompliano 🌪
｜ @APompliano
｜ 影响力: 244.0万粉丝 · 互动分 858.38
💡 核心观点: 作者引用 Brad Wilcox 与 Arthur Brooks 的观点，认为幸福的关键在于信仰、家庭、朋友和有意义的工作。
可信度: 30/100
事实核查:
- 无法核实: The key to happiness is having faith, family, friends, and meaningful work.（原文为引用观点并标注来源账号，未给出具体研究或数据。）
原文内容:
The key to happiness is having faith, family, friends, and meaningful work. (H/t @BradWilcoxIFS & @arthurbrooks) https://t.co/tW5ePzeL5l
⏰ 13:58 | ❤️ 595 点赞 | 📝 20词 | 入选: 高赞 595 | 查看原文 →
Prompter
｜ @PromptLLM
｜ 影响力: 9.8万粉丝 · 互动分 932.42
Prompter
｜ @PromptLLM
｜ 影响力: 9.8万粉丝 · 互动分 932.42
💡 核心观点: 作者提出生活质量与对不确定性的容忍度直接相关，未给出具体依据。
可信度: 18/100
事实核查:
- 无法核实: The quality of your life is in direct correlation to your tolerance to uncertainty（原文为断言式观点，未提供研究、数据或案例。）
原文内容:
The quality of your life is in direct correlation to your tolerance to uncertainty https://t.co/FugmUU3GiY
⏰ 21:24 | ❤️ 691 点赞 | 📝 17词 | 入选: 高赞 691 | 查看原文 →
Susan Zhang
｜ @suchenzang
｜ 影响力: 5.3万粉丝 · 互动分 347.4
Susan Zhang
｜ @suchenzang
｜ 影响力: 5.3万粉丝 · 互动分 347.4
💡 核心观点: 作者调侃在中国公园相亲角张贴约会头像打印件，让大爷大妈为你的婚恋前景操心，称之为「next-level chinamaxxing」。
可信度: 18/100
事实核查:
- 无法核实: put out a print of your dating profile pic at a people's park marriage market, where grand aunties and uncles can fret about your prospects（原文为个人生活化玩笑式描述，无可核验数据或来源。）
原文内容:
next-level chinamaxxing: put out a print of your dating profile pic at a people's park marriage market, where grand aunties and uncles can fret about your prospects https://t.co/S35frfqvCA
⏰ 08:01 | ❤️ 240 点赞 | 📝 32词 | 入选: 高赞 240 | 查看原文 →
Aakash Gupta
｜ @aakashgupta
｜ 影响力: 33.6万粉丝 · 互动分 7427.0
Aakash Gupta
｜ @aakashgupta
｜ 影响力: 33.6万粉丝 · 互动分 7427.0
💡 核心观点: 作者讲述2020年西班牙 MasterChef 选手 Saray Carrillo 端出未拔毛的生鸟被当场淘汰，并称另一选手 Andy 掌握分配权，只给她20分钟，而 Teresa 获75分钟。
可信度: 72/100
事实核查:
- 原文支持: He gave Teresa 75 minutes, Michael 65, Alberto 55, Iván 45 and José Mari 30. Saray got 20.（原文逐一列出各选手的烹饪时长，Saray 为20分钟，属具体可核验细节。）
- 无法核实: this was MasterChef Spain back in 2020（原文给出节目与年份，但未附节目片段或官方来源。）
- 无法核实: Beginners can burn 10 to 20 minutes plucking a regular chicken（该说法转述厨师 Hank Shaw 的经验，原文未给出研究或数据出处。）
原文内容:
A woman on MasterChef served the judges a raw, dead bird with every feather still on it and a few cherry tomatoes lined up down its back, and they threw her off the show on the spot. What most people never heard is that another contestant picked how long she got to cook. He gave her 20 minutes, and the woman at the other end of the line got 75. Her name is Saray Carrillo, this was MasterChef Spain back in 2020, and the night had been going sideways for her long before the bird showed up. She'd already been hit with the black apron in the first round after a meltdown over her cookies, and by the last challenge pretty much the whole cast had turned on her. So the show threw in a twist. Andy had won the previous round, which meant he got to decide who went into elimination and how much time each person got. He saved himself, dragged the people who'd already been safe back into the danger zone, and started handing out clocks. He gave Teresa 75 minutes, Michael 65, Alberto 55, Iván 45 and José Mari 30. Saray got 20. Her mystery box had a whole partridge in it, still wearing every feather, head and insides included. If you've never plucked a bird, it's slow and kind of miserable. Hank Shaw, a chef who cooks wild game for a living, says partridge and grouse skin tears so easily he does the breast one feather at a time. Beginners can burn 10 to 20 minutes plucking a regular chicken, and that's before you gut it or turn on a single burner. She didn't even try. She spooned a little bed of sauce onto the plate, combed the feathers flat with a knife, lined up the tomatoes, tucked it into some spring onions, and called it "dead bird on top of a plate." Jordi Cruz, one of the judges, told her she'd disrespected the 28,000 people who auditioned and that letting her through the door was the biggest mistake the show had made in eight years. Then he told her she didn't even deserve the black apron and to take it off. She said "bye bye" and danced her way out clapping. Afterward she said they wanted to watch her rip feathers out like a dog, and she wasn't going to give them that. She came back weeks later for a second-chance round, crying and apologizing, and didn't get back in. Two years after that she apologized to the show again in the middle of the night, saying she hadn't lived up to her contract. Andy kept going. About six weeks after the bird, he publicly apologized for deliberately sabotaging a different contestant and called his own behavior shameful. He finished second in the whole season.
引用 @poptime: E quando uma participante do MasterChef Espanha foi EXPULSA por servir um pássaro SEM LIMPAR 🤯 https://t.co/SwrlK3zgqi
⏰ 15:24 | ❤️ 6635 点赞 | 📝 469词 | 入选: 高赞 6635 / 引用讨论 / 长文 | 查看原文 →
Chubby♨️
｜ @kimmonismus
｜ 影响力: 15.8万粉丝 · 互动分 1264.84
Chubby♨️
｜ @kimmonismus
｜ 影响力: 15.8万粉丝 · 互动分 1264.84
💡 核心观点: 作者称Meta加入用AI推动科学发现的行列，并称除SpaceXSI外所有主要前沿实验室都在投入该方向。
可信度: 30/100
事实核查:
- 无法核实: Meta is joining the race for scientific breakthroughs using AI（原文给出该判断，引用内容仅为一句预告。）
- 无法核实: all major frontier labs except SpaceXSI（原文给出该范围表述，未列明实验室名单。）
原文内容:
Meta is joining the race for scientific breakthroughs using AI. As a result, all major frontier labs except SpaceXSI are now intensively working to advance scientific discovery. The competition will lead to unexpected breakthroughs here.
引用 @TomasMann1878: Wait till you see what we been up at @Meta 🙂
⏰ 18:36 | ❤️ 1018 点赞 | 📝 35词 | 入选: 高赞 1018 / 引用讨论 | 查看原文 →
其他
未归类的补充内容
Aakash Gupta
｜ @aakashgupta
｜ 影响力: 33.6万粉丝 · 互动分 6042.5
Aakash Gupta
｜ @aakashgupta
｜ 影响力: 33.6万粉丝 · 互动分 6042.5
💡 核心观点: 作者描述Faker通过F1至F5频繁切换镜头查看队友，并引用DeepMind星际争霸AI实验，说明受限视野会显著影响表现。
可信度: 78/100
事实核查:
- 原文支持: He's using five keys. F1 puts the camera on himself, F2 through F5 put it on each of his four teammates（原文明确写出按键与镜头对应关系。）
- 原文支持: Their first version could see the entire map at once and beat two pro players 10-0（原文给出具体比分与实验设定，属于可核验细节。）
- 无法核实: it lost its very first live game to a pro it had just swept 5-0（原文给出该说法，但未说明来源出处。）
原文内容:
Faker is 30 with six world championships, and his screen flashes like a strobe light because he won't leave his camera in one spot for more than a split second. Every flash is him jumping over to a teammate to see what's going on in their lane, then snapping back to his own fight. He does it over and over for the entire game. He's using five keys. F1 puts the camera on himself, F2 through F5 put it on each of his four teammates, and he taps Tab in between to pull up the scoreboard with everybody's items. In League your screen only shows a small circle of the map around your own champion, so this is how he sees the rest of it. The minimap in the corner shows little icons for where everyone is, which helps right up until you need to know how much health the enemy laner has left or how many minions are about to hit your tower. And he can't just check once. An enemy who showed up in top lane four seconds ago could be standing in the river by now, so whatever he saw is already going out of date by the time he looks back at his own lane. The only way to keep the whole map current in his head is to keep cycling, all game, while also playing mid against the best players in the world. Google's DeepMind found out how much this matters when they built an AI to play StarCraft, which has the same camera problem. Their first version could see the entire map at once and beat two pro players 10-0. When they rebuilt it to look through a camera like a human has to, it lost its very first live game to a pro it had just swept 5-0. To be fair, that camera version had only trained for a week. DeepMind kept at it, and for the version they published in Nature they made the camera mandatory, only let it act on what was on screen, and capped how fast it could click after checking with pro players. It still made Grandmaster, top 0.2% of the ladder. They basically ended up treating the camera as part of the game itself. OpenAI's Dota bots skipped the problem completely. They read the game straight from the API and never looked at a screen. Most AI agents that use a computer today are stuck with the keyhole too. They take a screenshot, figure out what to do, click, and take another screenshot. Faker runs his version of that loop faster than you can follow on video. He plays BLG on Sunday for a shot at his fifth straight Worlds final. Try watching his monitor for one teamfight and you'll get dizzy long before he does.
引用 @lolesports: Watching Faker play League: https://t.co/AGudIdi0XI
⏰ 08:37 | ❤️ 5274 点赞 | 📝 470词 | 入选: 高赞 5274 / 引用讨论 / 长文 | 查看原文 →
🔗 引用关系
- @aakashgupta 引用 @poptime：作者讲述2020年西班牙 MasterChef 选手 Saray Carrillo 端出未拔毛的生鸟被当场淘汰，并称另一选手 Andy 掌握分配权，只给她20分钟，而 Teresa 获75分钟。（原文：E quando uma participante do MasterChef Espanha foi EXPULSA por servir um pássaro SEM LIMPAR 🤯 https://t.co/SwrlK3zgqi）
- @aakashgupta 引用 @lolesports：作者描述Faker通过F1至F5频繁切换镜头查看队友，并引用DeepMind星际争霸AI实验，说明受限视野会显著影响表现。（原文：Watching Faker play League: https://t.co/AGudIdi0XI）
- @kimmonismus 引用 @TokenGremlin：作者称部分用户已被路由到GPT-6.2 Sol，并预测下周将是正式发布周，引用内容展示单提示生成Three.js场景的效果。（原文：GPT-6.2 Sol is already showing up for some ChatGPT users, and just look at what it's creating.
These aren't screenshots…）
- @GaryMarcus 引用 @0xFazerAI：作者引用Anthropic CEO关于经济年增10%、20%人口失业的预测，指出就业与生产率几乎未变，批评CEO随口给出数字。（原文：Anthropic's CEO described a future where the economy grows 10% a year and 20% of people don't have jobs.
In the video b…）
- @kimmonismus 引用 @TomasMann1878：作者称Meta加入用AI推动科学发现的行列，并称除SpaceXSI外所有主要前沿实验室都在投入该方向。（原文：Wait till you see what we been up at @Meta :)）
- @levelsio 引用 @apekshik：作者推荐独立开发者 @apekshik，称其作品都很酷，最新作品是一个无限演化的 Backrooms 模拟器，并附链接。（原文：i'm a solo dev and i built an infinitely evolving backrooms game and released it on the web.
35,000 players and ~9,600…）
- @swyx 引用 @skeptrune：作者认为 Cursor 被收购后表现强劲，称其2025年 ARR 达10亿美元、企业业务环比增长300%，并主张以 SEC 审计财报判断现实。（原文：@netspencer dead?）
- @levelsio 引用 @adamdotdev：作者回应称 AI 正在降低一切成本，先节省时间再降低周边开销，并重申会让所有东西便宜5到20倍。（原文：I’m at least 100x more productive than I was before AI. And that’s with a much better quality of life; I can get more do…）
🧩 跨推文归纳
共识
- 多位作者认同算力供给紧张将重塑定价与议价关系（2119409905434268064、2119518001150013823）。
- 关于 AI 冲击就业，不同作者援引相反证据，形成持续争论（2119408004764504458、2119465207940387031）。
分歧
- 一派认为 AI 泡沫破裂将带来多年高失业率，另一派指生产率与就业数据几乎未变，批评 CEO 随口给出数字。
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...