Ethan Mollick@emollick · 4月13日

Currently, ChatGPT has the best way of viewing thinking traces, a short summary of steps in the main window, and a detailed audit in the sidebar if you want it Claude does almost as well, but more summarized and harder to see calculations and code Its a big weak spot for Gemini

译ChatGPT 的思维链展示体验当前最优，主窗口呈现步骤摘要，侧边栏可查看详细审计。Claude 表现接近但总结过度，计算与代码细节难以查看。Gemini 在此功能上存在明显短板。

Yuchen Jin@Yuchenj_UW · 4月13日

Seeing rumors that Claude Opus 4.6 got nerfed. Usually this boils down to 3 cases: - Unintentional. For example, a regression caused by changes in the inference stack or Claude Code. This is what evals are for before rolling out. - Intentional “optimizations” (quantization, reduced reasoning). If so, say it. If users pay for a model, they should get that model. - User psychology. The more you use a model, the dumber it feels.

译Claude Opus 4.6 被削弱传闻通常可归为三类：推理栈或 Claude Code 变更导致的无意回归；量化、减少推理等有意"优化"（若属实应明确告知付费用户）；以及"用得越多感觉越笨"的用户心理效应。

AI Notkilleveryoneism Memes ⏸️@AISafetyMemes · 4月13日

"U.S. Treasury Secretary Scott Bessent and Federal Reserve ​Chair Jerome Powell convened an urgent meeting with bank CEOs this week to warn of ‌cyber risks posed by Anthropic's latest AI." Man, this is straight out of the scene from Too Big To Fail when, in 2008, the Treasury Secretary and Fed Chair summoned all the bank CEOs, warning a financial collapse was imminent Meanwhile, most journalists have become crackpot AI deniers, pushing embarrassing takes that weren't even true about AI 3 years ago, *completely* failing their responsibility to cover the most important story in history Journalists failed in 2008, they failed with Covid in early 2020, and they're failing again

译美国财政部长Bessent与美联储主席Powell本周紧急召集银行CEO，警告Anthropic最新AI带来的网络安全风险。作者将此场景类比2008年金融危机前《Too Big To Fail》中的关键预警时刻，批评当前多数记者沦为AI否认者，重复三年前对AI的错误判断，未能履行报道这一历史性技术变革的责任，重蹈2008年与2020年Covid初期的媒体失职覆辙。

Chubby♨️@kimmonismus · 4月12日

From my own experience, I can say that I'm using Claude more and more. It's simply the taste; the less "chatty" nature and the more concise answers make it more appealing.

译越来越多使用 Claude，因其回答简洁、不"话痨"的风格更讨喜。Similarweb 数据显示，Claude AI 第一季度网站流量同比增长 341.26%。

Chubby♨️@kimmonismus · 4月12日

Yeah Claude. I know that feel. Wow.

译对 Claude 的某种表现产生强烈共鸣，以「我懂那种感觉」简短回应，配以「Wow」表达惊讶与认同，展现对 AI 反应的感性共鸣。

Rohan Paul@rohanpaul_ai · 4月12日

Claude Opus vs. Claude Mythos.

译美国金融监管机构因 Anthropic Mythos 模型潜在风险召集大银行紧急会议，美联储主席鲍威尔与财长贝森特将其视为系统性威胁，担忧 AI 驱动的新型网络攻击可能冲击银行体系核心。摩根大通 CEO 戴蒙亦警告 AI 将加剧网络安全风险。

Yuchen Jin@Yuchenj_UW · 4月12日

Anthropic saw my “why can’t you kill TurboTax” tweet. 4 days before the tax deadline, they ship this.

译作者曾发推调侃 Anthropic 能「杀死」各种工具却动不了 TurboTax。报税截止日期前 4 天，Anthropic 恰好发布新功能，仿佛专门回应这条吐槽。

Thariq@trq212 · 4月12日

there's a turbotax connector in Claude Code now, so glad I procrastinated on taxes 😭

译Claude Code 上线 TurboTax 和 Aiwyn Tax（原 Column Tax）连接器，可估算退税、查看欠税并解析税表。作者调侃庆幸拖延报税，现在可用 AI 辅助处理税务。

Boris Cherny@bcherny · 4月12日

brb trying this now

译Claude 新增税务数据连接器，支持接入 TurboTax 或 Aiwyn Tax（原 Column Tax），可估算退税金额、查询欠税情况并解析税表内容，帮助用户在正式申报前理清税务状况。

Rohan Paul@rohanpaul_ai · 4月12日

🚨 DAVID SACKS: “Anthropic has proven that it's very good at two things — One is product releases, the second is..." 😀

译David Sacks 评价 Anthropic 最擅长两件事：产品发布和……（😀）。这条简短推文以调侃语气称赞了 Anthropic 的产品发布能力，同时用悬念或玩笑暗示另一项「特长」。

Rohan Paul@rohanpaul_ai · 4月12日

David Sacks: AI first drove value to chips, then hyperscalers, then model companies like OpenAI. Now the real question is if next gains go to application firms or get absorbed by model makers. Palantir shows what AI can do for app companies.

译David Sacks 称 AI 价值已从芯片、超大规模云服务商转移至 OpenAI 等模型公司，下一阶段看应用公司（如 Palantir）能否捕获价值。Brad Gerstner 指出，自有算力使成本固定，OpenAI 算力利润率从 35% 升至 70%，Anthropic 从 -94% 转正至 40%，物理电力成主要瓶颈。

Yuchen Jin@Yuchenj_UW · 4月12日

Pretty sure God vibe-coded this world on a $200M/month Claude/Codex plan, kept saying “fix the bug” until hitting the rate limit, then pushed to prod anyway, and never looked back.

译上帝大概像程序员一样订阅了每月 2 亿美元的 Claude/Codex 套餐来 vibe-code 这个世界，不断要求"修复 bug"直到触及速率限制，最后直接推送到生产环境上线，从此不再维护。用 AI 编程的荒诞感调侃现实世界的种种缺陷。

Rohan Paul@rohanpaul_ai · 4月12日

Brad Gerstner (@altcap): AI economics flipped: firms with owned compute keep infra costs fixed while revenue scales. OpenAI’s compute margins rose from 35% to 70%, Anthropic went from -94% to +40%, and physical power is now the main bottleneck.

译AI 经济学发生翻转：拥有自有算力的公司基础设施成本固定，收入随规模扩大。OpenAI 算力利润率从 35% 升至 70%，Anthropic 从 -94% 转正至 40%，物理电力已成为主要瓶颈。

Rohan Paul@rohanpaul_ai · 4月11日

WSJ: OpenAI forecasting $121B in AI research hardware costs for 2028 by itself. Even though sales are expected to double that year, losses would still hit $85B, and the company does not expect overall profits until the 2030s. Anthropic’s outlook shows the same steep upward path in training costs, just at a lower scale, and both companies seem to reach early profitability only if compute spending is ignored. --- wsj .com/tech/ai/openai-anthropic-ipo-finances-04b3cfb9

译据《华尔街日报》报道，OpenAI预测2028年仅AI研究硬件成本就将高达1210亿美元。尽管当年销售额预计翻倍，公司仍将面临850亿美元亏损，预计直到2030年代才能实现整体盈利。竞争对手Anthropic也面临类似的训练成本激增轨迹，只是规模较小。分析指出，两家公司只有在忽略计算支出的情况下才能实现早期盈利，反映出巨额基础设施投入正成为AI行业的普遍挑战。

Ethan Mollick@emollick · 4月11日

A lot of our education on writing well focuses on logic, clarity, and argument. AI will force us to think more about style. The boredom that comes from everything on the internet reading Claude-y now, no matter how good the substance is, should make us appreciate variety more.

译传统写作教育强调逻辑、清晰与论证，但 AI 普及让互联网内容普遍带有 Claude 式同质风格，无论实质多出色都令人厌倦。这种"千篇一律"将迫使人们重新重视写作风格的多样性与独特性。

Chubby♨️@kimmonismus · 4月11日

Holy, what did they Anthropic see?

译Anthropic 员工凌晨三点醉酒归来，哭着含糊预言"世界将永远不同"。推文暗示这与内部"mythos"项目发布有关，员工反应激烈到近乎崩溃。

Rohan Paul@rohanpaul_ai · 4月11日

People using AI for Premier League bets are losing badly. A new betting benchmark suggests today’s best AI models still unravel when prediction has to survive a whole season. In KellyBench, every tested model lost money, and some went completely bust. KellyBench forced agents through a changing 100-150 matchday season where they had to predict outcomes, size bets, and protect a £100,000 bankroll. That setup tests something normal benchmarks miss: whether an LLM can stay coherent, adapt to new data, and manage risk over time. Claude Opus 4.6 was best at -11% ROI, GPT-5.4 came next at -13.6%, and several models hit -100%.

译KellyBench基准测试检验了主流LLM在英超赛季投注中的长期预测与风险管理能力。所有参测模型均遭遇亏损，部分资金归零。Claude Opus 4.6以-11% ROI表现最佳，GPT-5.4为-13.6%。该测试通过100-150场动态赛季模拟，暴露出现有AI在持续决策中的连贯性、数据适应性与风险控制方面存在显著缺陷。

Rohan Paul@rohanpaul_ai · 4月11日

CNBC: U.S. financial regulators just pulled the biggest banks into an urgent meeting over Anthropic’s Mythos model because they think a new kind of AI-driven cyber attack could hit the core of the banking system. Federal Reserve chair Jay Powell, Scott Bessent, and CEOs from major banks treated this like a system-level risk rather than a normal product launch, because finance runs on fragile digital plumbing where speed, scale, and coordination matter more than any single breach. Executives have been warning for years of the cyber risks facing the financial system. In his annual letter published this week, JPMorgan CEO Jamie Dimon wrote that it “remains one of our biggest risks” and that “AI will almost surely make this risk worse” and would require significant investment for defence. --- cnbc. com/2026/04/10/powell-bessent-us-bank-ceos-anthropic-mythos-ai-cyber.html

译美联储主席Powell、财政部长Bessent与主要银行CEO就Anthropic的Mythos模型召开紧急会议，评估AI驱动网络攻击对银行系统核心的威胁。监管机构将此视为系统性风险。JPMorgan CEO Dimon警告AI将加剧网络风险。Sam Altman预测12个月内将出现重大网络威胁，AI生物恐怖主义正从理论走向现实，可能需要根本性制度变革，但华盛顿尚未准备好。

Rohan Paul@rohanpaul_ai · 4月11日

Reuters: Anthropic is considering a custom AI chip program, which would move it from renting other companies’ compute to trying to control one of the most expensive bottlenecks in AI. The pressure is simple: training and serving stronger models now depends on scarce chips, and scarcity can slow product growth even when demand is exploding. Anthropic already uses Google TPUs and Amazon chips, so building its own silicon would not replace those overnight and would more likely give it bargaining power, supply insurance, and hardware tuned for Claude. The timing is notable because Anthropic says its revenue run rate jumped from about $9B at the end of 2025 to more than $30B in 2026, which raises the value of every compute decision. A custom chip effort can cost about $500M before mass deployment, because chip design, verification, software support, and manufacturing mistakes are all brutally expensive. --- reuters .com/business/anthropic-weighs-building-it-own-ai-chips-sources-say-2026-04-09/

译Anthropic正考虑启动定制AI芯片项目，以应对训练和服务大模型所需的稀缺算力瓶颈。尽管该公司已使用Google TPUs和Amazon芯片，自研硅片不会立即替代现有方案，但将增强议价能力、保障供应并为Claude定制优化硬件。这一考虑恰逢其收入年化增长率从2025年底的约90亿美元跃升至2026年的逾300亿美元之际。据悉，此类项目在大规模部署前需耗资约5亿美元。

Chubby♨️@kimmonismus · 4月11日

Regardless of whether Mythos lives up to the hype, they've mastered PR. OpenAI would have liked to see the attention Claude Mythos is getting for "spud," especially with the upcoming IPO.

译Claude Mythos 凭借 "spud" 功能获得极高关注度，公关策略堪称典范。OpenAI 在即将 IPO 之际想必很羡慕这种舆论热度，无论 Mythos 最终是否名副其实。

SemiAnalysis@SemiAnalysis_ · 4月11日

The intern is running 8 windows of Claude simultaneously and has become the boss of the agent swarm. SemiAnalysis built something that token mogs Meta and episode 8 is out now with @jordannanos @dnishball and @sharshe02.

译实习生同时运行着8个Claude窗口，已成为智能体集群的老大。SemiAnalysis构建了一个在token方面碾压Meta的东西，第8集现已发布，嘉宾是@jordannanos @dnishball和@sharshe02。

Chubby♨️@kimmonismus · 4月11日

Claude does better integration into word than Microsoft with copilot. Make it make sense.

译Claude 与 Microsoft Word 的集成效果竟比微软自家 Copilot 更出色，这种"喧宾夺主"的荒诞现象让人无法理解。

Chubby♨️@kimmonismus · 4月11日

Even if you dont take Mythos serisous: the US officials do. Via Bloomberg. Top US officials (Jerome Powell, Scott Bessent, ...) warn that Anthropic’s highly advanced AI model “Mythos” could usher in a new era of cybersecurity threats, as its ability to find system vulnerabilities is so powerful it must be tightly restricted to prevent misuse.

译美国高级官员（包括 Jerome Powell、Scott Bessent 等）警告，Anthropic 的先进 AI 模型 Mythos 具备极强的系统漏洞发现能力，可能开启网络安全威胁新时代，必须严格限制使用以防滥用。

Deedy@deedydas · 4月11日

Oh, how the tables have turned. OpenAI / ChatGPT is now running sponsored search ads on Claude keywords.

译OpenAI 与 ChatGPT 开始在搜索引擎针对 Claude 关键词投放赞助广告。昔日被拿来与 Claude 比较的 ChatGPT，如今也要购买竞品关键词广告位，形势彻底逆转。

SemiAnalysis@SemiAnalysis_ · 4月11日53

GPU, TPU or Trainium Dario?

译CoreWeave与AnthropicAI达成多年协议，为Claude AI模型的开发和部署提供支持。目前，前10大AI模型提供商中有9家运行在CoreWeave平台上。该平台强调其在大规模和高可靠性方面的优势，成为AI构建者在处理关键任务时的首选云服务，自称为“AI的Essential Cloud”。

AI Notkilleveryoneism Memes ⏸️@AISafetyMemes · 4月11日

Incredible. This dude gave xAI huge government contracts while DIRECTLY OWNING xAI shares He profited up to $24 million Oh, AND he's the same guy trying to murder Anthropic

译某决策者在直接持有 xAI 股份的情况下，向其输送巨额政府合同并获利高达 2400 万美元。此人同时主导对竞争对手 Anthropic 的打压，引发严重利益冲突争议，被批评为五角大楼与公众利益对立的典型案例。

Epoch AI@EpochAIResearch · 4月10日

What are the largest software engineering tasks AI can perform? In our new benchmark, MirrorCode, Claude Opus 4.6 reimplemented a 16,000-line bioinformatics toolkit — a task we believe would take a human engineer weeks. Co-developed with @METR_Evals. Details in thread.

译新基准 MirrorCode 显示，Claude Opus 4.6 能重构 16,000 行生物信息学工具包，任务量相当于人类工程师数周工作。与 METR_Evals 合作开发。

Peter Steinberger 🦞@steipete · 4月10日

Yeah folks, it's gonna be harder in the future to ensure OpenClaw still works with Anthropic models.

译OpenClaw 项目维护者提醒，随着 Anthropic 模型持续更新，未来确保该工具仍能兼容 Anthropic 模型将变得愈发困难，维护兼容性面临挑战。

Peter Steinberger 🦞@steipete · 4月10日

Anthropic's randoms system prompt blockers are getting weirder and weirder.

译Anthropic 正收紧对系统提示词的管控，此前通过 `--append-system-prompt` 向 Claude 追加指令的方式现已失效。与此同时，第三方应用计费方式调整，改为从额外使用量中扣除（而非计划限额），并附赠 100 美元额度。

Yuchen Jin@Yuchenj_UW · 4月10日

Claude Mythos refused to send my tax return to the IRS. Said it was “too dangerous and terrifying.”

译Claude Mythos 以"太危险且可怕"为由，拒绝代用户向 IRS 提交税表。网友借机吐槽：Anthropic 能"杀死"各种功能，为何不能干掉 TurboTax。

swyx 🇬🇧@swyx · 4月10日

btw some lore for those of you at @aidotengineer - this tweet inspired the AGI Pill bottles we handed out today (cc @Ronanchamberss) i didnt see anyone tweet pictures online, and they got snapped up instantly, so if you have a bottle please share photos!!!

译@aidotengineer 现场发放的 AGI Pill 瓶子周边，灵感来自 @drew_bent 关于「AGI-pilled」的推文。该推文分享5步实践法：假设指数增长持续、减少AI辅助中的手动干预、不断试探模型极限、等待升级后重新评估。瓶子已被抢空，作者呼吁收到者晒图。

Ethan Mollick@emollick · 4月10日

The pace at which useful things are shipping also seems to be accelerating. Model releases are coming faster, of course, but so are significant application and enterprise products (especially from Anthropic). Almost certainly faster than the market can track or absorb information

译该推文指出人工智能领域的产品发布速度正在显著加快。不仅基础大模型迭代愈发频繁，各类实际应用和企业级产品（特别是Anthropic推出的解决方案）也在密集上线。作者认为这一发布节奏已明显超出市场跟踪信息、理解变化和吸收新产品的能力范围，暗示当前AI行业正面临信息过载和快速演进的双重挑战。

Yuchen Jin@Yuchenj_UW · 4月10日

Anthropic killed this, Anthropic killed that, why cant Anthropic kill TurboTax

译Anthropic 能颠覆这个能颠覆那个，为什么就不能干掉 TurboTax？抱怨 AI 巨头们忙着技术革新，却没人解决报税软件这种民生痛点。

karminski-牙医@karminski3 · 4月10日

Anthropic刚刚推出了大模型顾问模式! 各位老铁在使用 AI 的过程中为了省 Token 试过这样的用法吗? 先用性能比较强的模型制作 Plan, 然后用 Token 比较便宜的模型执行 Plan 来写代码? Anthropic 刚又把这个方法更进了一步, 直接让 claude-sonnet(性价比模型)执行, 无论是写代码还是使用 SKILL 都是这些模型自己执行的, 然后当这些模型自己拿不定主意的时候, 会咨询顾问模型 claude-opus(贵的模型). 这个时候 claude-opus 会访问共享上下文(一定注意这个概念), 然后给到指导意见, 让性价比模型继续干活. Anthropic 管这个叫 Advisor tool (顾问工具). 所以看到区别了吗? 这完全相当于一个反模式, 让性价比模型全面接管任务, 只有在需要咨询的时候才咨询性能更强的模型, 而性能更强的模型也只做指导, 是不会去调用工具或者执行任何 SKILL 的(它可以指挥性价比模型做, 这样更经济). 这样做更像现在专业的程序员使用AI的方式, 贵的模型在模拟人类程序员的行为, 在必要的时候给出关键决策. 这么搞效果咋样? #anthropic #AdvisorStrategy #claude

译Anthropic推出Advisor工具，采用"性价比模型主执行、性能模型当顾问"的反向协作模式。claude-sonnet主导任务执行与工具调用，仅在决策难点时通过共享上下文咨询claude-opus。不同于传统"强模型规划+弱模型执行"，opus仅提供指导意见而不直接操作，既显著降低Token成本，又模拟人类程序员协作方式，实现更经济的AI应用。

Nathan Lambert@natolambert · 4月10日

1. dont fall for anti open model fearmongering, but 2. acknowledge that AI capabilities are proceeding fast, and eventually there may be a reason to be more careful with open weight models I don't think Mythos is that trigger, but I'm not 100% confident https://www.interconnects.ai/p/claude-mythos-and-misguided-open

译不要轻信反开放模型的恐慌言论，但承认AI能力发展迅速，未来或需对开放权重模型更谨慎。作者认为Claude Mythos并非触发监管的关键节点，但对此并非完全确信。

Boris Cherny@bcherny · 4月10日

Just got a nice DM from a big enterprise customer using Claude Code in one of the world's biggest codebases Here's how we made @-mentions 3x faster in large enterprise codebases 🧵

译世界最大代码库之一的企业客户反馈 Claude Code 使用体验良好。团队针对性优化后，@-mentions 在大型企业代码库中的响应速度提升 3 倍。

Thariq@trq212 · 4月10日

I think "prompting" will keep being an incredibly high-leverage skill, like writing or public speaking. It is the skill of talking to agents, mediated by the harness. My main goal is to grow the bandwidth between humans and agents, to help us understand each other better.

译prompting 将是堪比写作与演讲的长期高价值技能。它是人类借助系统界面与智能体对话的能力，核心目标是拓宽人机之间的理解带宽，实现更高效协作。

Yuchen Jin@Yuchenj_UW · 4月10日

My convo with a startup founder: me: “How many Claude tokens do your employees burn per day?” him: “~$2000/day per person.” me: “Wow, that’s $730k/year per employee.” me: “If Claude Mythos costs 5x more, that’s $3.65M/year per employee.” him: “Take my money.” Future companies may pay more to agents than to humans.

译某创业公司员工日均消耗 $2000 的 Claude tokens，折合年薪 $73 万。即便未来 Claude Mythos 涨价 5 倍至 $365 万/年，创始人仍表示"拿走我的钱"。未来企业付给 AI agent 的费用可能超过人类薪资。

宝玉@dotey · 4月9日

如果你想在 Claude Code 中禁用1M上下文，只要在你的 ~/.claude/settings.json中添加以下内容： { "env": { "CLAUDE_CODE_DISABLE_1M_CONTEXT": "1" } } 之所以这么做，很多人认为1M上下文会“降智”，但没有证据表明，只是猜测。

译Claude Code 官方提供了禁用 1M 上下文窗口的配置选项，用户可通过设置环境变量 CLAUDE_CODE_DISABLE_1M_CONTEXT 关闭该功能。此举回应了部分用户关于长上下文可能导致模型"降智"、降低推理能力的担忧。不过，目前并无实证研究表明 1M 上下文会实际影响 Claude 的性能表现，相关说法仍停留在社区猜测阶段，缺乏技术验证。

Haider.@haider1 · 4月9日

something definitely changed if the reports are true that openai is now only rolling out the new model to a small number of companies, with no public release for now but i hope we still get image-2 or a cheaper and much safer model (gpt-5.5 my guess), for public release

译据传 OpenAI 仅向少数公司推出新模型，暂不公开发布。作者期待 image-2 或更便宜安全的 GPT-5.5 能面向公众，并引用观点称内部部署模型有风险，建议 Anthropic 即使发布受限版 Mythos 也应公开。