【AI 英文奏折】07月07日

x每日奏折14小时前发布 tianming

7 0 0

【AI 英文奏折】2026年07月07日

共收录 21 篇深度内容

📋 今日内容速览

快速浏览，点击感兴趣的推文查看详细分析

Anthony Pompliano: 跨党派支持的儿童投资账户计划获政商界力推。
Ethan Mollick: 明确目标比技巧更能提升AI效果
Machina: Karpathy预见的自写解题书模型现已实现。
Rohan Paul: 核心观点总结中…
Rohan Paul: 核心观点总结中…
Rohan Paul: 机器人手部精细动作或导致按摩行业裁员。
Gary Marcus: 核心观点总结中…
Santiago Valdarrama: 核心观点总结中…
Rohan Paul: 久坐超30分钟增癌症风险，轻度活动可缓解。
Google Gemini: 核心观点总结中…
Google Gemini: 核心观点总结中…
Google Gemini: Gemini Spark可实时追踪话题并分析赛事。
Rohan Paul: 核心观点总结中…
Sakana AI: Sakana AI推出多语种实时翻译校对工具。
Rohan Paul: 核心观点总结中…
Machina: Fable5将停用，需提前提取其智能到文件中保存。
Emily: 核心观点总结中…
clem 🤗: 核心观点总结中…
SemiAnalysis: 核心观点总结中…
Charly Wargnier: 核心观点总结中…
Marc Lou: DataFast bot traffic新增页面分组、搜索和爬虫文件追踪功能。

📖 详细内容

Anthony Pompliano @apompliano

Entrepreneur, investor, and lifelong learner. | 影响力: 2134k万粉丝

💡 核心观点: 跨党派支持的儿童投资账户计划获政商界力推。

可信度: 10/10 – 2项声明可直接验证；2项需进一步确认；1项为观点陈述

事实核查:

◐ 部分可验证: 该推文作者今日参与了椭圆形办公室的“Invest America / Trump Accounts”倡议启动活动 (可通过白宫公开日程或媒体报道验证作者是否出席活动，但需确认具体人员名单和活动内容。)
◦ 观点: 该计划可能拥有作者所见过的两党支持度最高的 bipartisan support（两党支持） (两党支持程度是主观判断，缺乏量化数据（如投票记录或公开声明）的直接支持。)
✓ 可验证: 戴尔家族（Dell Family）为该计划捐赠了超过60亿美元 (可通过慈善机构或项目官网的公开捐赠记录验证，或通过戴尔家族及项目方的正式声明确认。)

原文内容:

I was in the Oval Office today for the launch of the Invest America / Trump Accounts initiative. 

Here are my takeaways:

1. This program may have the most bipartisan support of anything I have ever seen. Everyone, regardless of political party, understands the power of giving young children money to compound over decades.

2. While the government and current administration should get immense credit for the program, there were many private citizens like @altcap, @MichaelDell and others that persistently pursued the idea until it became reality. They quite literally changed the course of millions of lives with this idea.

3. It will quickly become a game of status for people to donate to fund the accounts. The Dell Family gave over $6 billion. SpaceX’s @Gwynne_Shotwell gave more than $300 million today. Brad Gerstner is adopting every child in the state of Indiana to give money to their accounts. The list goes on and on already…

4. Companies will begin making contributions core parts of their perk packages for employees. Corporations like Bank of America, JPMorgan, Robinhood and others have already announced these efforts.

5. The Oval Office is surprisingly small in size, but the historical significance of the room is palpable the second you walk into it. The President called it “the most important room in the world” today and I walked away thinking that was accurate.

6. This idea of Invest America accounts has become a magnet for the world’s most successful people. I saw Larry Fink, David Solomon, Lisa Su, and many other Fortune 500 CEOs in attendance for the Rose Garden lunch. It is hard for an idea to fail when this many people are behind it.

7. The Rose Garden was very nicely done. I had been to the White House before, but never to the Oval Office or the Rose Garden. Both seemed elevated compared to the rest of the White House and surrounding buildings. It sounds like the new ballroom will also be a modern structure that mirrors this elevated approach.

8. The staff and admin team in the White House are generous and genuine. Each person was kind, gracious, and seemed to be at the top of their game. Sometimes you go to places and are disappointed with the let down from your expectations, but this was not the case with the White House team. All professionals doing their job.

9. The media loves Trump way more than they let on publicly. Many of the mainstream hosts and anchors were joking around with him like it is one big club, while simultaneously laughing at his jokes and being much more friendly than I expected. He seems to be good for their business, so it makes sense why the public narrative and private interactions don’t appear to match.

10. I never thought I would see a day where @NYSE and @Nasdaq collaborated on an event like this, including the ringing of both bells by one person at the same time. I guess anything is possible if kids are involved. 

Overall, this was an awesome experience. I know some people like this administration and some people don’t. I learned from @kevinolearytv to focus on policies, not politicians. 

And the Invest America accounts are objectively a great idea that will hopefully spread like wildfire in the years to come.

⏰ 02:36 | ❤️ 3476点赞 | 📝 543词 | 查看原文 →

↑ 返回顶部

Ethan Mollick @emollick

Professor @Wharton studying AI, innovation & startups. Democratizing education using tech. Author of Co-Intelligence | 影响力: unknown万粉丝

💡 核心观点: 明确目标比技巧更能提升AI效果

可信度: 4/10 – 1项需进一步确认；2项为观点陈述

事实核查:

◐ 部分可验证: Even before the agentic revolution, prompting tricks stopped being very valuable, as our research has shown. (该声明提到“研究显示提示技巧价值下降”，但未提供具体研究数据或来源。若作者所属机构公开了相关研究（如论文、报告），则可验证；否则需依赖第三方复现或实测。)
◦ 观点: The best approach to AI right now is to clearly specify your goals, your output, what “good” & bad look like, how to test the results… (该声明属于方法论建议（如“最佳实践”），是主观观点而非客观事实。其有效性可能因任务或模型而异，缺乏统一验证标准。)
◦ 观点: (yes, this is just management) (括号内内容是对前文的比喻性总结（将AI提示比作管理），属于主观解读，无客观依据可验证。)

原文内容:

Even before the agentic revolution, prompting tricks stopped being very valuable, as our research has shown.

The best approach to AI right now is to clearly specify your goals, your output, what "good" & bad look like, how to test the results... (yes, this is just management)

⏰ 09:41 | ❤️ 219点赞 | 📝 46词 | 查看原文 →

↑ 返回顶部

Machina @exm7777

running ai-powered agencies | weeklyaiops.com | 影响力: unknown万粉丝

💡 核心观点: Karpathy预见的自写解题书模型现已实现。

可信度: 10/10 – 2项声明可直接验证；3项需进一步确认

事实核查:

◐ 部分可验证: Karpathy一年前曾提出模型自我编写问题解决手册的概念，但当时未被关注 (需查找Karpathy公开演讲、文章或社交媒体记录以确认是否提及此概念，但“未被关注”属于主观判断，难以量化验证。)
◐ 部分可验证: Fable 5是当前最智能的模型，且其Claude订阅功能将于今日结束 (需核实Fable 5是否为公开模型及其性能排名（如基准测试），但“最智能”缺乏标准定义；订阅截止时间可通过官方渠道验证。)
✓ 可验证: 通过5个工作流程（如“plant its standards”）可提取模型知识并生成可执行手册 (流程描述具体但未提供实际案例或公开工具验证效果，依赖用户实测结果，当前信息不足。)

原文内容:

kinda insane that Karpathy talked about this a year ago but no one paid attention...

he described a model writing a book for itself on how to solve problems... but that concept is real now

and today is the last day to have the smartest model write yours: Fable 5 leaves every Claude subscription tonight

here's how you run the extraction with 5 workflows:

1. plant its standards

open each project and have fable rewrite your claude[.]md as the operating manual a weaker model would need: your conventions, the mistakes to prevent with the rule that stops each one, the quality bar as checkable criteria
then have it write your 3 highest-leverage skills in full

2. the consultant audit

give it your projects and numbers with one prompt: "act as the consultant i can't afford, audit everything, deliver a roadmap a less capable model can execute step by step
the reasoning gets written down while the model that produced it is still flat-rate"

3. the second brain run

fire deep research runs on your niche, your competitors, your customers' problems
mine every run into an Obsidian vault, one insight per note... a hundred linked notes get reused, one long report gets buried

4. /goals + dynamic workflows

describe what done looks like, demand the proof pasted in the finish line, cap every run with a turn or time limit... the model builds unattended for hours while the tokens still cost nothing extra

5. the recorder skill

one skill that fires after every hard problem: the approach, the judgment calls, the reusable rule, saved as a note in your repo
wire it into your claude rules and fable documents its own thinking all day, automatically

check out the full guide:

⏰ 09:44 | ❤️ 95点赞 | 📝 285词 | 查看原文 →

↑ 返回顶部

Rohan Paul @rohanpaul_ai

Compiling in real-time, the race towards AGI. The Largest Show on X for AI. | 影响力: 0万粉丝

💡 核心观点: 核心观点总结中…

可信度: 10/10 – 3项声明可直接验证；2项需进一步确认

事实核查:

✓ 可验证: By 2025, Anthropic had already identified a half-trillion-dollar coding market. (该声明涉及Anthropic内部未公开的市场预测数据，缺乏官方公开报告或第三方权威来源支持，无法直接验证。)
◐ 部分可验证: Within Anthropic at that time (early 2025), people were spending $100 a day on tokens, comes out to $20,000 or $30,000 a year. (若Anthropic公开其用户消费数据或提供API定价历史，可部分验证；但“$100/天”是否为典型用户行为需更多样本支持，目前依赖单一信源。)
◐ 部分可验证: There are around 20 million coders in the world, leading to a half-trillion-dollar market from coding alone. (全球开发者数量（20 million）可通过行业报告（如IDC、Statista）间接验证，但“半万亿美元市场”的计算依赖假设（如人均支出），需结合具体模型评估。)

原文内容:

By 2025, Anthropic had already identified a half-trillion-dollar coding market.

"Within Anthropic at that time (early 2025), people were spending $100 a day on tokens, comes out to $20,000 or $30,000 a year.

If you think about how many coders there are in the world, around 20 million, you have got a half-a-trillion-dollar market just from coding alone. And hat was based on 9-month-old technology."

Alex Sacerdote, the Founder of Whale Rock Capital Management, talking about early 2025,  when Claude Code started exploding.

----

From "Invest Like The Best" YouTube channel, (link in comment)

⏰ 09:36 | ❤️ 22点赞 | 📝 92词 | 查看原文 →

↑ 返回顶部

Rohan Paul @rohanpaul_ai

Compiling in real-time, the race towards AGI. The Largest Show on X for AI. | 影响力: 0万粉丝

💡 核心观点: 核心观点总结中…

可信度: 10/10 – 3项声明可直接验证；2项需进一步确认

事实核查:

◐ 部分可验证: Frontier AI in healthcare has a hidden failure mode: it can look medically brilliant while being clinically unready. (需查阅《Nature Medicine》发表的原始研究以确认具体实验设计和结论，但推文未提供直接链接或论文标题，需进一步检索验证。)
✓ 可验证: AI models in the study were brittle, giving correct答案 in normal tests但失败于轻微修改的测试（如问题调整、关键信息删除或图文设置变化）。 (若找到原始论文，可通过研究方法（如压力测试设计）和结果数据直接验证此声明。但当前缺乏直接论文引用，需依赖用户自行检索。)
◐ 部分可验证: 部分模型在关键输入被删除后仍能猜出正确答案，可能依赖捷径而非真正理解医学案例。 (需分析论文中的具体案例和模型行为（如输入消融实验），但推文未提供细节，验证需依赖完整研究。)

原文内容:

《自然-医学》发表的这项研究对医疗人工智能发出严厉警告。

尖端医疗AI存在隐性缺陷模式：它可能在医学层面表现卓越，却远未达到临床应用标准。

研究者在健康基准测试中评估前沿AI模型后，又通过压力测试检验这些模型是真正稳健，还是仅擅长应试。

结果发现模型存在脆弱性。

即模型在常规测试中能给出正确答案，但当问题微调、关键信息缺失或图文设置改变时就会出错。

一个反常现象是：即使移除关键输入，某些模型仍能猜中正确答案，这表明它们可能在使用捷径而非真正理解医学案例。

这些模型有时会给出听起来专业且合乎逻辑的解释，但其推理过程存在缺陷。

最终结论并非"AI在医学领域无用"，而是"基准测试的成功不等同于临床可用性"。

⏰ 13:38 | ❤️ 61点赞 | 📝 168词 | 查看原文 →

↑ 返回顶部

Rohan Paul @rohanpaul_ai

Compiling in real-time, the race towards AGI. The Largest Show on X for AI. | 影响力: 0万粉丝

💡 核心观点: 机器人手部精细动作或导致按摩行业裁员。

可信度: 6/10 – 1项声明可直接验证；2项需进一步确认；1项为观点陈述

事实核查:

◐ 部分可验证: Xynova将在2026年维也纳ICRA会议上展示机器人手 (ICRA（国际机器人与自动化会议）官网可查询未来会议日程及参展方，但2026年信息可能尚未完全公开，需后续确认Xynova的参展信息。)
◐ 部分可验证: 该机器人手具备协调的手指动作（握拳、伸展、闭合等）
◦ 观点: 该机器人手将导致按摩店裁员 (这是对技术社会影响的推测，无客观数据支持，属于主观观点。)

原文内容:

这款机械手将导致按摩行业出现裁员潮。  
协调的指关节运动——握拳、伸指、张合自如。完整的手部闭合动作。掌心开合与精准捏取控制，尽在2026年维也纳国际机器人与自动化会议（ICRA）上的Xynova展台。  

（注：根据中文科技新闻惯例，"digit control"译为"指端控制"更符合专业语境，但考虑到与前后动作描述的连贯性，采用"精准捏取控制"的意译方式；"ICRA 2026"按国际会议命名规范补充完整称谓并保留英文缩写，符合国内科技媒体报道标准）

⏰ 12:50 | ❤️ 93点赞 | 📝 35词 | 查看原文 →

↑ 返回顶部

Gary Marcus @garymarcus

OG GenAI Skeptic; spoke at US Senate. Advocating world models | 影响力: unknown万粉丝

💡 核心观点: 核心观点总结中…

可信度: 6/10 – 1项声明可直接验证；1项需进一步确认；1项为观点陈述

事实核查:

◐ 部分可验证: The Treasury Department reportedly knows that the massive GenAI build out poses systemic risks to the US financial system. (该声明提到“据报道”（reportedly），但未提供具体来源。可通过查找美国财政部公开报告、听证会记录或权威媒体报道验证，但需确认信息来源的可靠性。)
✓ 可验证: The Treasury Department doesn’t want to acknowledge the risks publicly. (涉及机构“不愿公开承认”的动机属于推测，缺乏直接证据（如官方声明或内部文件），目前无法验证其真实性。)
◦ 观点: People may talk about this moment for years. (这是对未来的主观预测，无客观事实依据，属于个人观点或推测。)

原文内容:

 Wow. The Treasury Department reportedly knows that the massive GenAI build out poses systemic risks to the US financial system – and doesn’t want to acknowledge it publicly.  

People may talk about this moment for years.

⏰ 08:42 | ❤️ 260点赞 | 📝 36词 | 查看原文 →

↑ 返回顶部

Santiago Valdarrama @svpino

| 影响力: unknown万粉丝

💡 核心观点: 核心观点总结中…

可信度: 4/10 – 2项需进一步确认；1项为观点陈述

事实核查:

◐ 部分可验证: Fable优化后，技能平均缩短40%且功能保持不变 (需实测对比优化前后的技能代码或性能数据，但无公开基准测试或第三方验证，依赖用户自行确认。)
◐ 部分可验证: Fable通过并行写入使某技能（基于MCP服务器）运行速度提升2倍 (需访问具体技能代码及MCP服务器日志以验证并行写入的实现和速度提升，但缺乏公开技术文档或案例支持。)
◦ 观点: “The model is pretty freaky!”（Fable模型效果惊人） (属主观评价，无客观性能指标或对比数据支持。)

原文内容:

I maintain several skills. I run regularly with Claude Code.

I asked Fable to read every skill and optimize them.

They are now 40% shorter on average! Everything works as before, but Fable cut out a ton of fat.

One of the skills writes on a spreadsheet using an MCP server. Fable found a way to write multiple rows in parallel, and the skill runs twice as fast as it used to.

If you still don't know how to use Fable, use it to optimize something you already have. The model is pretty freaky!

⏰ 20:30 | ❤️ 115点赞 | 📝 94词 | 查看原文 →

↑ 返回顶部

Rohan Paul @rohanpaul_ai

Compiling in real-time, the race towards AGI. The Largest Show on X for AI. | 影响力: 0万粉丝

💡 核心观点: 久坐超30分钟增癌症风险，轻度活动可缓解。

可信度: 10/10 – 3项声明可直接验证

事实核查:

✓ 可验证: New University of Glasgow led study finds, sitting longer than 30 minutes at once linked to higher cancer death risk. (可通过检索格拉斯哥大学官网或相关学术期刊发布的原始研究论文验证该声明。研究方法和结论通常会在论文中详细说明，属于公开可获取的科学数据。)
✓ 可验证: Same study also says even light activity like just ironing may ease risks tied to long periods of sitting. (同一研究的结论应可在原始论文或官方新闻稿中找到支持。具体活动（如熨烫）与风险缓解的关联性需查看研究中对“轻度活动”的定义和数据分析部分。)
✓ 可验证: 该研究由格拉斯哥大学主导（University of Glasgow led study）。 (可通过研究论文的作者 affiliations 或格拉斯哥大学的公开研究项目列表确认其主导地位。)

原文内容:

格拉斯哥大学最新主导的研究发现，单次静坐超过30分钟与癌症死亡风险升高存在关联。

该研究同时指出，即便是熨衣服这样的轻度活动，也可能缓解长时间静坐带来的健康风险。

⏰ 12:35 | ❤️ 27点赞 | 📝 38词 | 查看原文 →

↑ 返回顶部

Google Gemini @geminiapp

The Gemini app turns research into reality, bringing frontier AI experiences like Veo 3.1, Deep Think, Nano Banana, and more to hundreds of millions of people. | 影响力: 488.10k万粉丝

💡 核心观点: 核心观点总结中…

可信度: 8/10 – 2项声明可直接验证；1项需进一步确认

事实核查:

✓ 可验证: Gemini Spark目前以Beta版本向美国Google AI Ultra订阅用户开放 (可通过官方链接（http://gemini.google/overview/agent/spark）或Google AI订阅页面直接验证功能开放范围和用户权限)
◐ 部分可验证: 团队正努力在本月晚些时候将Gemini Spark推广至更多国家 (需后续观察实际发布情况，当前仅能通过官方声明间接验证其计划，但无法确认具体时间或国家列表)
✓ 可验证: Gemini Spark是Google AI的一项新功能（或产品） (推文附带的官方链接和品牌名称（Gemini Spark）可直接关联到Google AI的公开信息)

原文内容:

Gemini Spark is currently available in Beta for Google AI Ultra subscribers in the U.S., and our team is working hard to bring Gemini Spark to more countries later this month.

Learn more about Gemini Spark here: http://gemini.google/overview/agent/spark…

⏰ 07:53 | ❤️ 44点赞 | 📝 44词 | 查看原文 →

↑ 返回顶部

Google Gemini @geminiapp

The Gemini app turns research into reality, bringing frontier AI experiences like Veo 3.1, Deep Think, Nano Banana, and more to hundreds of millions of people. | 影响力: 488.10k万粉丝

💡 核心观点: 核心观点总结中…

可信度: 8/10 – 2项声明可直接验证；1项需进一步确认

事实核查:

◐ 部分可验证: Gemini Spark可以通过指定提示词自动发送定制化球队比赛邮件 (需实测验证是否在Gemini平台存在”Spark”功能模块，且能否通过输入指定提示词实现邮件自动生成和发送。Google官方文档未明确提及此具体功能。)
✓ 可验证: 访问http://gemini.google或应用可找到Spark功能 (可通过直接访问链接或检查官方应用确认是否存在该入口，但”Spark”是否为公开功能需进一步验证（当前Gemini官网未显示此命名模块）。)
✓ 可验证: 提示词“Email me a match summary…”能触发自动化邮件服务 (无公开证据表明Gemini支持通过自然语言指令创建自动化邮件工作流，需实际测试且结果可能受账户权限/地区限制影响。)

原文内容:

Here’s how to use Gemini Spark to set up custom emails after your favorite team plays:
1) Open http://gemini.google or the app
2) Tap “Spark” in the side menu
3) Insert the following prompt: “Email me a match summary and analysis after each [INSERT TEAM] soccer match.”

⏰ 07:53 | ❤️ 36点赞 | 📝 47词 | 查看原文 →

↑ 返回顶部

Google Gemini @geminiapp

The Gemini app turns research into reality, bringing frontier AI experiences like Veo 3.1, Deep Think, Nano Banana, and more to hundreds of millions of people. | 影响力: 488.10k万粉丝

💡 核心观点: Gemini Spark可实时追踪话题并分析赛事。

可信度: 8/10 – 2项声明可直接验证；1项需进一步确认

事实核查:

◐ 部分可验证: Gemini Spark can now intelligently track topics and react to events in real time. (需实测或查看官方功能演示以确认其“智能追踪”和“实时反应”的具体能力，但若官方提供技术文档或案例，可部分验证。)
✓ 可验证: Try the prompt in the next post to get tailored game analyses emailed to you after your favorite team plays. (可通过实际测试推文中的提示（prompt）功能，或检查官方是否提供该服务的说明页面/用户反馈来直接验证。)
✓ 可验证: Gemini Spark provides tailored game analyses. (需验证分析内容的个性化程度（如是否基于用户指定球队），但若官方公开样本报告或用户案例，可部分验证。)

原文内容:

Gemini Spark can now intelligently track topics and react to events in real time. 

Try the prompt in the next post to get tailored game analyses emailed to you after your favorite team plays:

⏰ 07:52 | ❤️ 437点赞 | 📝 34词 | 查看原文 →

↑ 返回顶部

Rohan Paul @rohanpaul_ai

Compiling in real-time, the race towards AGI. The Largest Show on X for AI. | 影响力: 0万粉丝

💡 核心观点: 核心观点总结中…

可信度: 10/10 – 3项声明可直接验证；2项需进一步确认

事实核查:

◐ 部分可验证: LLMs只能处理离散的符号序列，而语言是对世界的简化描述 (该声明基于Yann LeCun对LLM技术原理的解释，可通过自然语言处理（NLP）领域的学术文献部分验证，但“语言是对世界的简化描述”属于认知科学范畴的概括性观点，需结合多学科研究综合判断。)
✓ 可验证: 最大规模的LLMs预训练数据量为20万亿单词（30万亿token，约10¹⁴字节） (当前主流LLM（如GPT-4、PaLM）的训练数据量可通过技术论文或企业白皮书（如OpenAI、DeepMind公开资料）查证，但具体数值可能因模型版本和统计口径存在差异。)
✓ 可验证: 四岁儿童通过视觉获得的数据量相当于10¹⁴字节，而阅读同等文本需40万年 (儿童感官数据量与文本阅读时间的对比涉及神经科学估算（如人类视觉信息传输速率）和假设性计算（如连续阅读速度），目前缺乏权威统一的量化标准。)

原文内容:

During a Bloomberg interview, Yann LeCun (@ylecun ) explains why LLMs are limited in terms of real-world intelligence during a Bloomberg interview.

"Language is a very approximate, reduced, quantized, and simplified description of the world, and LLMs can only deal with discrete sequences of symbols. The world is much more complicated than language.

The biggest LLMs are pre-trained on the totality of all the publicly available text on the internet. That’s about 20 trillion words, or 30 trillion tokens.

A token is about 3 bytes. So total 10¹⁴ bytes of text.

This is the amount of data a four-year-old has seen through vision during four years. Now, the text, though, would take 400,000 years to read?

So, there is enormously more data from sensory input, like vision, touch, and everything else, than there could ever be through language."

A child does not need 400,000 years of reading to understand cups, doors, balance, faces, falls, or heat, because the body is already collecting dense feedback from vision, touch, motion, and consequence.

Text strips most of that away.

It turns a living scene into symbols, then asks the model to infer the missing world from traces left by people describing it.

That is why an LLM can sound fluent about physics and still have no native sense of how fragile glass feels in a hand.

Moravec’s paradox names this reversal: the things humans find intellectual can be easier for machines than the things toddlers do without applause.

The hard part is not producing an answer, but building a model of the world that survives contact with weight, friction, surprise, and failure.

----

Link to the full video on Bloomberg's site. Link in comment.

⏰ 14:04 | ❤️ 152点赞 | 📝 280词 | 查看原文 →

↑ 返回顶部

Sakana AI @sakanaailabs

Building Frontier AI in Japan | Try Sakana Chat, Marlin, Fugu → sakana.ai | 影响力: unknown万粉丝

💡 核心观点: Sakana AI推出多语种实时翻译校对工具。

可信度: 10/10 – 2项声明可直接验证；2项需进一步确认；1项为观点陈述

事实核查:

◐ 部分可验证: Sakana Translate支持日语、英语和中文的实时长文本翻译 (可通过官网链接（https://translate.sakana.ai）测试翻译功能，但“实时长文本”的具体表现（如最大字数限制、延迟等）需实测确认。)
✓ 可验证: Sakana Translate提供校对功能，优化语气和措辞，并显示修改痕迹 (官网若明确展示校对界面或提供试用示例（如带修订标记的文本对比），可直接验证；否则需注册/试用确认具体功能。)
◐ 部分可验证: Sakana Translate提供“询问”功能，可澄清词汇的细微用法 (需实测验证该功能是否存在及如何运作（如是否内置问答模块），但官网若提供功能说明或截图则可直接验证。)

原文内容:

We released our in-house Japanese, English, and Chinese translation tool, Sakana Translate!

Try it→ https://translate.sakana.ai 

• Translate: Handles long text in real time
• Proofread: Tone and phrasing refined, with tracked changes
• Ask: Nuanced word choices clarified

⏰ 00:00 | ❤️ 186点赞 | 📝 39词 | 查看原文 →

↑ 返回顶部

Rohan Paul @rohanpaul_ai

Compiling in real-time, the race towards AGI. The Largest Show on X for AI. | 影响力: 0万粉丝

💡 核心观点: 核心观点总结中…

可信度: 8/10 – 2项声明可直接验证；1项需进一步确认

事实核查:

◐ 部分可验证: Grok 4.5基于xAI的V9基础模型，参数规模1.5T (模型参数规模通常由官方公布，但需等待xAI或马斯克等直接确认；若仅依赖“多方报道”，则属间接信息。)
✓ 可验证: V9预训练于2026年5月26日完成 (未来日期（2026年）的声明无法验证，且无官方来源或历史规律佐证，可能为推测或笔误（如应为2024年）。)
✓ 可验证: xAI计划在2026年底前每月发布全新训练模型 (此类计划属公司愿景或目标，无具体执行证据，且“每月发布”的可行性存疑，需后续观察实际动作验证。)

原文内容:

Grok 4.5 almost ready to drop?

Some details about Grok 4.5 that have now been confirmed by various reports.

- Grok 4.5 is built on xAI’s V9 foundation model with 1.5T parameters.

- That makes it about 3x larger than v8-small, which currently serves production traffic on X.

- V9 pre-training was completed on May 26, 2026.

- Cursor coding data was added after pre-training through supplemental training.

- Reinforcement learning is still ongoing.

- xAI is planning monthly new-model releases trained from scratch through the end of 2026.

- Grok 4.3 remains the current public API model on Amazon Bedrock.

⏰ 05:25 | ❤️ 60点赞 | 📝 87词 | 查看原文 →

↑ 返回顶部

Machina @exm7777

running ai-powered agencies | weeklyaiops.com | 影响力: unknown万粉丝

💡 核心观点: Fable5将停用，需提前提取其智能到文件中保存。

可信度: 1/10 – 1项暂无法验证

事实核查:

✗ 无法验证: 事实核查功能暂时不可用 (系统处理中)

原文内容:

仔细听我说，因为知道这个的人不多...

Fable 5明天将退出所有Claude订阅服务，但在该模型本身无法使用后，仍有办法保留其智能

诀窍在于提取：让Fable将其判断写入文件，任何更便宜的模型都能运行

以下是今天必须启动的5个工作流程：

> 将它的标准植入你的工作区：Claude文件 + 以Fable级别编写的技能，永远由Opus执行  
> 顾问审计：Fable通读你的整个业务并撰写路线图，更便宜的模型只需遵循  
> 第二大脑运行：将深度研究挖掘到金库中，供未来每次会话读取  
> /目标 + 动态工作流：在代币仍是固定费率时，进行数小时无人值守的构建  
> 记录器技能：Fable今天解决的每个难题都会被记录，这样它的思考就会留在你的仓库中  

文章中已写出所有提示，可直接粘贴使用：

⏰ 00:50 | ❤️ 416点赞 | 📝 149词 | 查看原文 →

↑ 返回顶部

Emily @iamemily2050

Any sufficiently advanced technology is indistinguishable from magic. | 影响力: 48.2k万粉丝

💡 核心观点: 核心观点总结中…

可信度: 1/10 – 1项暂无法验证

事实核查:

✗ 无法验证: 事实核查功能暂时不可用 (系统处理中)

原文内容:

我希望看到更多人表达对人工智能发展及其未来的看法，因为这项技术将彻底改变我们的生活。总会有人不喜欢、不认同甚至发表恶意评论，但不要让这些消极的声音影响你的思考或压制你的观点。

⏰ 06:39 | ❤️ 22点赞 | 📝 48词 | 查看原文 →

↑ 返回顶部

clem 🤗 @clementdelangue

Co-founder & CEO @HuggingFace 🤗, the open and collaborative platform for AI builders | 影响力: 405.4k万粉丝

💡 核心观点: 核心观点总结中…

可信度: 1/10 – 1项暂无法验证

事实核查:

✗ 无法验证: 事实核查功能暂时不可用 (系统处理中)

原文内容:

This is how how much data AI builders are storing on HF Xet (replaced git storage fully in ~Nov 25). Feels like this is just the beginning and should get to exabytes soon!

⏰ 06:03 | ❤️ 52点赞 | 📝 32词 | 查看原文 →

↑ 返回顶部

SemiAnalysis @semianalysis_

In-depth research on semiconductors, AI infra & hardware | 影响力: unknown万粉丝

💡 核心观点: 核心观点总结中…

可信度: 10/10 – 2项声明可直接验证；2项需进一步确认；1项为观点陈述

事实核查:

◐ 部分可验证: Nvidia GPU Debt Backstop Unleashes the AI Project Trinity: Capital, Offtake and Datacenters (需核实Nvidia是否正式提出“AI Project Trinity”及其具体内容，部分信息可能来自非官方解读或行业分析。)
✓ 可验证: Over 7T AI debt by 2029 (该数据缺乏明确来源或计算方法，可能为预测或估算，无法通过公开渠道直接验证。)
✓ 可验证: Nvidia’s Objective is to Broaden Compute Access (Nvidia官网或公开演讲中多次提及扩大计算资源可及性的目标，可通过官方声明验证。)

原文内容:

Nvidia GPU Debt Backstop Unleashes the AI Project Trinity: Capital, Offtake and Datacenters
Over 7T AI debt by 2029, 
There can be no Neoouds without the Trinity. 
Nvidia's Backstop Economics Explained. 
AI Debt Needs Quantified. 
Nvidia's Objective is to Broaden Compute Access

⏰ 05:57 | ❤️ 106点赞 | 📝 42词 | 查看原文 →

↑ 返回顶部

Charly Wargnier @datachaz

Ex @Streamlit @Snowflake Maestro • I write about AI agents, LLMs and automation • My ❤️ is open source • DM for collabs | 影响力: unknown万粉丝

💡 核心观点: 核心观点总结中…

可信度: 1/10 – 1项暂无法验证

事实核查:

✗ 无法验证: 事实核查功能暂时不可用 (系统处理中)

原文内容:

 THE "PIRATE BAY" FOR OPEN LLMS IS HERE

Hugging Bay is a new verified artifact registry for downloading model weights via torrents and hosted mirrors.

But it’s far more than just a file dump.

It’s a beautifully structured catalog where every artifact is checked for provenance and license clarity before indexing.

You can easily filter across categories:
→ LLMs & Embeddings (EMB)
→ Visual (VIS) & Audio (AUD) models
→ Open-source Agents & Datasets
→ Apps, Tools, and Evals

.. and there are many more feats to explore.

They’ve also built in semantic rerank.

It means you can search for natural-language queries like "best small commercial embedding model for RAG" and actually get a clean result.

really cool stuff.

check out the registry below ↓

⏰ 14:19 | ❤️ 120点赞 | 📝 121词 | 查看原文 →

↑ 返回顶部

Marc Lou @marclou

⭐️ TrustMRR.com $27K/m | DataFa.st $20K/m | SHlPORDIE.COM $20K/mo | 影响力: 170.0k万粉丝

💡 核心观点: DataFast bot traffic新增页面分组、搜索和爬虫文件追踪功能。

可信度: 1/10 – 1项暂无法验证

事实核查:

✗ 无法验证: 事实核查功能暂时不可用 (系统处理中)

原文内容:

I added 2 upgrades to DataFast bot traffic 

You can now:

 See bot traffic grouped by page

So instead of only seeing “ChatGPT crawled my site”, you can see which exact URLs bots request most.

 Search crawled pages

Useful for checking if AI/search bots requested pages like /pricing, /docs, /llms.txt, etc.

 Track crawler discovery files (NPM package update needed)

The server-side tracker now keeps requests to:

- robots.txt
- llms.txt
- sitemaps

This matters more now because AI agents and crawlers use these files to understand what exists on your site.

Would love feedback if you’re tracking this kind of traffic already.

PS. Bot traffic tracking is included in existing DataFast plans.

⏰ 20:47 | ❤️ 96点赞 | 📝 114词 | 查看原文 →

↑ 返回顶部

# x每日奏折