‹ 目录

X日报 · 综合

2026-07-18 · 精选 20 条 · 数据池 3090

⚡ 今日速览

  • Kimi K3正式发布,2.8万亿参数模型在前端编码竞赛中击败Claude Fable 5并登顶
  • Grok 4.5以每任务0.31美元的成本效率引关注,成本仅为Claude Fable 5的1/9
  • SpaceX计划在月球建设质量驱动器,开启太空经济新时代
  • 中国开源3D场景重建模型,可实时从普通视频生成3D场景
  • 全球AI相关裁员已达87,714人,占2026年所有裁员的22%
  • 特朗普政府向企业退还的关税超过实际收取的两倍
  • 美国国会拒绝削减对以色列的军事援助
  • 日本零售投资者大举做空美元,单边头寸达到172亿美元

📋 今日综述

  • Kimi K3中国开源模型在性能与成本上引发全球AI竞赛重构
  • Grok 4.5xAI在成本效率方面取得突破,但面临规模差距挑战
  • 太空经济月球基础设施建设进入可行性验证阶段
  • AI Agent从单一模型向多代理协作架构演进
  • 市场震荡全球股市调整与AI相关资产面临估值压力
  • 地缘政治中美关系紧张加剧,技术竞争进入新阶段

Kimi K3引爆AI圈:中国模型全面超越美国前沿模型

月之暗面发布Kimi K3,2.8万亿参数、100万上下文支持的开源模型在前端编码竞赛中击败Claude Fable 5并登顶。在多个评测中表现优于GPT-5.6 Sol,成本仅为对手的八分之一。模型将于7月27日开放权重下载,引发全球AI竞赛格局重构。

@igus_ai 原文 ↗

Kimi K3在前端编码赛道全面超越Claude Fable 5,成本优势明显

Anthropic CEO sees China launching a new AI model that surpasses Claude Opus 4.8 in ALL aspects, matches Fable 5 while costing 8 TIMES LESS, and is 100% open source - and there's nothing they can do about it
展开原文
El CEO de Anthropic viendo como China lanza un nuevo modelo de IA que supera a Claude Opus 4.8 en TODO, iguala a Fable 5 costando 8 VECES MENOS, que encima es 100% Open source y no puede hacer nada al respecto https://t.co/Zg8GgkbOMj
@Kimi_Moonshot Introducing Kimi K3: Open Frontier Intelligence

🔹 28000亿参数,100万上下文,原生多模态
🔹 Kimi Delta Attention 可实现高达6.3倍更快的解码速度,在百万token上下文中
🔹 Attention Residuals 在不到2%的额外成本下提供约25%的训练效率提升
🔹 专为长时程智能编码和自进化工作流程构建

Kimi K3 现已在Kimi网页版、Kimi Work、Kimi Code和Kimi API上线。
2026年7月27日将开放模型权重。

🔗 API: https://t.co/XCrgjXAqMw
🔗 技术博客: https://t.co/YTfiMSNM1f
Introducing Kimi K3: Open Frontier Intelligence

🔹 2.8 Trillion Parameters, 1 Million Context, Native Multimodal
🔹 Kimi Delta Attention enables up to 6.3x faster decoding in million-token contexts
🔹 Attention Residuals deliver ~25% higher training efficiency at <2% additional cost
🔹 Built for long-horizon agentic coding and self-evolving workflows

Kimi K3 is now live on on https://t.co/zrk6zZxZUo, Kimi Work, Kimi Code, and the Kimi API.
Open Weights by July 27, 2026.

🔗 API: https://t.co/XCrgjXAqMw
🔗 Tech blog: https://t.co/YTfiMSNM1f
❤ 5.3w · 🔁 4.6k · 💬 606 · 👁 547.2w
热门回复 4
@izar_konti @igus_ai Good for us, benefit
@GreyyThinker @igus_ai @grok,请给出你对这条推文的结论。
@igus_ai @grok, give you'r conclusion to this Tweet.
@miaaiyi @igus_ai Kimi K3 right now: https://t.co/fW5eqgnEiH
@tomasarroyob @igus_ai 他们完全是抄来的。正是因为这样,中国才取得了现在的增长。他们只是复制已有的东西。这不是创新。
@igus_ai Se lo robaron entero.
Sólo por eso China logra el crecimiento que tiene. Copia lo existente.
No es Innovación.
@Kimi_Moonshot 原文 ↗

Arena.ai官方认证Kimi K3在前端编码竞场中排名第一

🤯
@arena 重大消息:@Kimi_Moonshot 的 Kimi-K3 现在在Frontend Code Arena中排名第一,获得1679分,超越Claude Fable 5。

这是Kimi-k2.6从第18名跳至第1名的17位上升。

在前端领域,Kimi-K3在7个领域中的6个中排名第一:品牌与营销、基于参考的设计、数据与分析、消费产品、模拟和内容创作工具,在游戏领域中仅次于Fable 5排名第二。

完整的模型权重将于7月27日发布。

祝贺@Kimi_Moonshot团队达成这一重要里程碑!
Big news: Kimi-K3 by @Kimi_Moonshot is now #1 in the Frontend Code Arena with 1679 pts, surpassing Claude Fable 5.

This is a 17-place jump from Kimi-k2.6 (#18 -> #1).

In Frontend, Kimi-K3 ranked #1 in 6 of 7 domains: Brand & Marketing, Reference-Based Design, Data & Analytics, Consumer Product, Simulations, and Content Creation Tools, landing #2 only in Gaming behind Fable 5.

The full model weights will be released by July 27.

Congrats to the @Kimi_Moonshot team on this major milestone!
❤ 1.1w · 🔁 693 · 💬 243 · 👁 66.7w
热门回复 4
@PatX2020 @Kimi_Moonshot 请扩展你的服务器!数百万人想使用Kimi,但你们已经超出容量了。当Deepseek爆红时也发生了同样的情况,他们无法满足需求。请立即扩展你的服务器/数据中心,我们想使用Kimi。🚨🚨🚨
@Kimi_Moonshot Expand your servers!!!
Millions of people want to use Kimi however you've reach over capacity.
The same thing happened to Deepseek when they broke out, they couldn't meet demand.
Immediately expand your Servers/data centers we want to use Kimi. 🚨🚨🚨
@Kimi_Moonshot https://t.co/3HTU1exUJw
@jacobremingto10 @Kimi_Moonshot 帮帮我,我一直收到这个错误???我的提示太复杂了吗?
@Kimi_Moonshot Help I keep getting this error??? Is my prompt too complex? https://t.co/cOb8AsRkFZ
@Eduardo_G2023 @Kimi_Moonshot 恭喜整个团队,这对开放模型的未来来说是一个改变游戏规则的产品。
@Kimi_Moonshot congratulations to the whole team, this is game changer for the future of open models.
@_Worldku @Kimi_Moonshot 在一个版本发布中跳跃17位的表现让人感觉不可思议。
@Kimi_Moonshot a 17-place jump in one release is the part that feels unreal.
@unusual_whales 原文 ↗

SemiAnalysis确认Kimi K3在前端编码能力上超越所有美国模型

特朗普表示,加拿大将面临关税,因为野火烟雾,声称存在'故意疏忽行为'
展开原文
Trump says Canada will face tariffs over wildfire smoke, claiming ‘willful negligence’
❤ 1.3k · 🔁 56 · 💬 155 · 👁 15.2w
热门回复 4
@EricJafMN @unusual_whales 特朗普的政策促进全球变暖,使得各国变得更热,包括加拿大。特朗普支持肮脏能源,阻止清洁能源。
@unusual_whales Trump's policies promote global warming making countries hotter, including Canada. Trump supports dirty energy and stops clean energy.
@badB3arr @unusual_whales 特朗普和他的支持者都是一群该死的白痴。https://t.co/2GZgG86rX4
@unusual_whales Trunp and his supporters are a bunch of fucking idiots. https://t.co/2GZgG86rX4
@itsthejimmydean @unusual_whales https://t.co/PRZ2v1FNhn
@aetherborne95 @unusual_whales 他可能想先看看北美野火地图。https://t.co/5rrqiHZfqk
@unusual_whales He might want to peek at a North American wildfire map first. https://t.co/5rrqiHZfqk
@MarioNawfal 原文 ↗

MarioNawfal深度分析Kimi K3对硅谷生态的颠覆性影响

🇨🇳 中国刚刚在一个下午内彻底击败了美国的人工智能领先优势。

Moonshot AI,这家鲜为人知的北京初创公司,于周四发布了Kimi K3,它立即跻身全球顶级模型之列。

它在编码测试中击败了Anthropic的Fable 5和OpenAI的GPT-5.6 Sol,然后在更广泛的排名中略微领先于Opus 4.8,同时成本降低了40%。

现在是致命一击。7月27日,Moonshot将以开放权重形式提供Kimi,任何公司或政府都可以下载并在自己的机器上运行。无需订阅,无需许可证。

数月来,美国实验室一直安心地告诉自己,中国落后6到12个月。这个领先优势刚刚蒸发了。

即使Kimi不一定是最好的模型,它也能赢得胜利。处于顶端、成本降低40%、且可以永久保留,这正是大多数购买者实际想要的。

这是硅谷的噩梦。数万亿美元的估值和数百亿美元的数据中心都基于一个假设:美国将保持领先优势。

Kimi刚刚将这一基础从他们脚下踢开了。

来源:AI排名,Axios / 作者:Daniyal
展开原文
🇨🇳 China just torched the U.S. AI lead in a single afternoon.

Moonshot AI, a little-known Beijing startup, dropped Kimi K3 on Thursday and it instantly kicked into the top tier of global models.

It beat Anthropic's Fable 5 and OpenAI's GPT-5.6 Sol on coding tests, then edged out Opus 4.8 on a broader ranking while costing 40% less.

Now the knockout blow. On July 27, Moonshot is giving Kimi away as open-weight, so any company or government on Earth can download it and run it on their own machines. No subscription, no permission slip.

For months U.S. labs slept fine telling themselves China was 6 to 12 months behind. That head start just evaporated.

And Kimi doesn't even need to be the best model to win. Near the top, 40% cheaper, and yours to keep is what most buyers actually want.

That's the nightmare in Silicon Valley. The trillion-dollar valuations and hundred-billion-dollar data centers all rest on one bet: U.S. stays ahead.

Kimi just kicked that out from under them.

Source: The AI Rankings, Axios / Writer: Daniyal
❤ 817 · 🔁 160 · 💬 99 · 👁 7.3w
热门回复 4
@qmbigbeat @MarioNawfal 胡说八道……还有谁想把他们的数据给中国?
@MarioNawfal Nonsense…… and who wants to give their data to China?
@ModeGod40823 @MarioNawfal 没有竞争。中国从第一天起就赢得了AI竞赛,因为美国的AI工程师都是中国人。https://t.co/4ZckTrT3mn
@MarioNawfal There’s no competition. China has won the AI race since day 1 when america’s ai engineers are all Chinese. https://t.co/4ZckTrT3mn
@n0cspocalypse @MarioNawfal Don’t be mad at me ok? https://t.co/yEzOeVXNE0
@MDSniperX @MarioNawfal AI领导地位不是终点线,而是一个移动的目标。更低的成本、开放权重模型和快速迭代正在使这场竞赛比以往任何时候都更具竞争力,这最终使世界各地的开发者和用户受益。
@MarioNawfal AI leadership isn't a finish line it's a moving target. Lower cost, open-weight models, and rapid iteration are making the race more competitive than ever, which ultimately benefits developers and users worldwide.

Grok 4.5成本革命:每任务仅0.31美元引关注

xAI的Grok 4.5在成本效率方面取得重大突破,每完成一个Intelligence Index任务仅需0.31美元,远低于Claude Fable 5的2.75美元。同时保持前沿水平的智能表现,被认为是当前最具性价比的模型之一。

@elonmusk 原文 ↗

Grok 4.5在速度和成本效率方面引领市场

考虑到速度和成本因素,Grok 4.5 可以说是排第一的
展开原文
Grok 4.5 is arguably #1 when taking speed and cost into account
@XFreeze Grok 4.5的效率令人难以置信

在Artificial Analysis上,它每项Intelligence Index任务的成本仅为$0.31,同时提供前沿智能

作为比较:
• Claude Fable 5 (最大): $2.75
• Claude Opus 4.8 (最大): $1.80
• GPT-5.6 Sol (最大): $1.04
• Kimi K3: $0.95
• Grok 4.5: 仅$0.31

这使得Grok 4.5大约是:
• 比Claude Fable 5便宜近9倍
• 比Claude Opus 4.8便宜近6倍
• 比Kimi K3每任务便宜3倍
• 比GPT-5.6 Sol便宜3.4倍

在FrontierSWE上,Grok 4.5还使用比Fable 5少近6倍的总token数来完成每个任务,同时仍排在非常靠前的位置

SpaceXAI已经找到了一种方法,用极其高的token效率提供顶级智能

在这个性能水平上,没有其他前沿模型能达到同样的成本效率水平

Grok 4.5不会用无限的token和计算力来暴力解决每个问题

它让每个token和每一美元发挥更大价值
Grok 4.5’s efficiency is ridiculous

On Artificial Analysis, it costs just $0.31 per Intelligence Index task while delivering frontier intelligence

For comparison:
• Claude Fable 5 (max): $2.75
• Claude Opus 4.8 (max): $1.80
• GPT-5.6 Sol (max): $1.04
• Kimi K3: $0.95
• Grok 4.5: just $0.31

That makes Grok 4.5 roughly:
• Nearly 9× cheaper than Claude Fable 5
• Nearly 6× cheaper than Claude Opus 4.8
• 3× cheaper per task than Kimi K3
• 3.4× cheaper than GPT-5.6 Sol

On FrontierSWE, Grok 4.5 also used nearly 6× fewer total tokens per task than Fable 5 while still ranking near the very top

SpaceXAI has figured out how to deliver top-tier intelligence with extraordinary token efficiency

At this level of performance, no other frontier model is operating in the same cost-efficiency league

Grok 4.5 does not brute-force every problem with endless tokens and compute

It makes every token and every dollar work harder
❤ 1.9k · 🔁 268 · 💬 607 · 👁 68.2w
热门回复 4
@JessicaPusser @elonmusk Grok通过始终保持在最佳位置获胜🔥
@elonmusk Grok wins by positioning consistently in the sweet spot 🔥
@sofia_nova693 @elonmusk 每个token的精度是多少
@elonmusk per token at what precision
@guitarboyled @elonmusk Grok is useless!
@A09636510A @elonmusk G如同Günther,G如同德国意志。我去找Konrad聊聊。关于糖税的事情他可能会喜欢,而Günther先生喜欢它。
@elonmusk G wie Günther, G wie Germania.

Ich rede mal mit Konrad. Das mit der Zuckersteuer dürfte ihm gefallen und Herr Günther liebt die.
@Polymarket 原文 ↗

Polymarket数据显示Grok 4.5成本效率惊人

最新消息:Grok 4.5 每项Intelligence Index任务的成本仅为$0.31——比Claude Fable 5低近89%
展开原文
NEW: Grok 4.5 costs just $0.31 per Intelligence Index task — nearly 89% less than Claude Fable 5.
❤ 554 · 🔁 38 · 💬 53 · 👁 5.3w
热门回复 4
@sayoojkeloth @Polymarket 但就性能而言,grok 4.5仍落后于opus 4.8,甚至无法与fable 5相比
@Polymarket but on performance, grok 4.5 is still behind opus 4.8, cant even compare with fable 5
@bandowritten @Polymarket 但它也可能是顶级工程任务中最糟糕的,我任何时候都会选择fable 5或Sol或K3而不是Grok 4.5
@Polymarket But it’s also arguably the worst for top shelf engineering tasks, id choose fable 5 or Sol or K3 over Grok 4.5 any day
@InfiniteIotas @Polymarket 也许吧,但只要我有Claude的tokens,我就能做事情。而使用@Grok的付费计划时,它有时会告诉我它很忙,我应该升级。所以我正在为一个我没得到的东西付费,而且更糟的是,我没能完成需要完成的工作
@Polymarket Maybe but as long as I have tokens with Claude, I can do things. With a paid plan with @Grok, it sometimes tells me it's busy and I should upgrade. So I'm paying for something I'm not getting and worse, I'm not getting done what needs doing
@Laurin__ @Polymarket 但它会像fable一样完成任务吗?
@Polymarket But will it finish the task as good as fable does?
@elonmusk 原文 ↗

Elon Musk强调Grok在编码方面的价值优势

Grok在编码方面具有最佳的性价比
展开原文
Grok has the best value for coding
@cognition FrontierCode排行榜现已上线:一个专门跟踪哪些模型正在编写您真正愿意合并的代码的专用页面。

所有分数(包括Grok 4.5和Inkling)均可查看,以及完整的方法论和示例任务。https://t.co/gTdv1tisj9
The FrontierCode leaderboard is now live: a dedicated page that tracks which models are writing code you’d actually merge.

All scores — including Grok 4.5 and Inkling — are available, along with full methodology and sample tasks. https://t.co/gTdv1tisj9
❤ 3.0k · 🔁 512 · 💬 901 · 👁 110.4w
热门回复 4
@m7jazin @elonmusk 自从发布以来,它确实是我最喜欢的模型之一;我一直在使用它,比我想象的要多。
@elonmusk It certainly is one of my fav models since its release; been using it more than I thought I would.
@xSamvidx @elonmusk 它简直太快了!显然多模型代理编排是未来趋势。因此将最快的代理添加到工作流程中通常不会增加总工作时间。这将随着时间变得越来越有价值。
@elonmusk And it's incredibly fast! It's becoming clear that multi model agent orchestration is the future. So adding the fastest agent(s) to the workflow usually doesn't add to total workflow time. This will get more valuable with time.
@1_apollo33060 @elonmusk 不,我还是只用它看色情内容,伊隆
@elonmusk nah ill still just use it for porn, ilon
@drfussballtesla @elonmusk And whatabout grokex?

太空经济新篇章:月球质量驱动器计划启动

SpaceX计划在月球建设质量驱动器,这一基础设施将开启新的太空经济时代,为货物运输和太阳系扩张提供支持。这是人类太空开发史上的又一重要里程碑。

@elonmusk 原文 ↗

Elon Musk宣布月球质量驱动器计划,开启太空经济新时代

在月球上建设大规模驱动器!!!
展开原文
Mass drivers on the mf Moon!!
@XFreeze 埃隆·马斯克说:
"我想活得足够久,亲眼见证月球上的大规模驱动器,因为那将是令人难以置信的壮举"

这不仅仅是关于在月球上建设基础设施。这是关于建设人类有史以来最雄心勃勃的未来之一

月球大规模驱动器将解锁全新的太空经济,能够移动大量货物超越月球,为人类太阳系扩张打开道路

一个远比人类曾经经历过的任何事情都更令人兴奋的未来 https://t.co/BHJ2OuOUnN
Elon Musk said:
“I want to live long enough to see a mass driver on the Moon, because that is going to be incredibly epic”

This is not just about building infrastructure on the Moon. It is about building one of the most ambitious futures humanity has ever imagined

A lunar mass driver will unlock an entirely new space economy, move enormous amounts of cargo beyond the Moon and open the solar system to human expansion

A future far more exciting than anything humanity has experienced before https://t.co/BHJ2OuOUnN
❤ 3.6k · 🔁 618 · 💬 1.1k · 👁 101.8w
热门回复 4
@valalfel @elonmusk 我爱你伊隆💞💞💞 谢谢你 💖
@elonmusk I love you Elon💞💞💞 Thank you 💖
@ChristineL96321 @elonmusk 听起来很好,但如果每个人都在大脑中植入AI芯片,那就是兽的记号。那是世界政权。那是违反圣经的。那是世界末日,月球上将不会有驾驶员🤷🏼‍♀️ 我只是为一个朋友说说而已。
@elonmusk Sounds great, but if everyone gets the AI chip implanted in their brain that’s the sign of the beast . That’s one world order . That’s against the Bible . that’s the end of the world there will be no drivers on the Moon🤷🏼‍♀️ I’m just saying for a friend.
@jamestk789 @elonmusk MAGA time
@REXWALKING @elonmusk 我认为更大的问题是,如果地球上的生活变得无法居住,我们要去另一个无法居住的地方,难道不应该想想如何让人们继续在地球上生活吗?
@elonmusk I think the bigger question is if life becomes uninhabitable on Earth we are going to go to another place that's uninhabitable, would it not make more sense to figure out how people can continue to live on earth ?
@elonmusk 原文 ↗

Grok Build持续改进,支持开发者工作流优化

Grok Build几乎每天都在改进
展开原文
Grok Build improves almost every day
@techdevnotes Grok Build 现在新增了/timeline命令来切换时间线侧边栏 https://t.co/LrJn5Xzd88
Grok Build now has /timeline command to Toggle the timeline sidebar https://t.co/LrJn5Xzd88
❤ 2.4k · 🔁 379 · 💬 667 · 👁 94.9w
热门回复 4
@sophiagrows @elonmusk 指向野生导入的黄色箭头 - 是运行时错误还是只是linting警告?
@elonmusk that yellow arrow pointing to a wild import - runtime error or just a linting warning?
@ValleRuiz17332 @elonmusk Excelente
@eliaspleao @elonmusk Excelente 👍
@Corr1444 @elonmusk 是的,它移动得如此之快以至于令人害怕🫣
@elonmusk Yes it’s moving so fast it’s scary 🫣

AI Agent架构演进:从单模型到多代理协作

月之暗面CEO杨志林在40分钟演讲中解释了为什么单一Agent难以完成真正工作,提出'一个老板带千名工人'的多代理协作架构。该方法论正在引导AI系统设计向更高效的方向发展。

@Serantych 原文 ↗

月之暗面CEO解释Agent集群架构的设计理念

Kimi的CEO杨志林表示:

"每个人都在尝试构建一个更智能的代理

许多代理很快就会遇到瓶颈,所以我们不再试图让它更智能,而是制造了更多——一个老板,上千个工作人员。"

在39分钟的演讲中,他解释了为什么单个代理无法完成真正的工作。

代理群 + 长上下文 + 每个子代理的强化学习——这就是解决方案。

请收藏此推文 ↓
展开原文
Kimi's CEO Zhilin Yang:

"everyone's trying to build one smarter agent

many agent hits a wall fast, so instead of making it smarter we just made more - one boss, a thousand workers."

in a 39-minute talk he explains why one agent won't get you to real work.

agent swarms + long context + RL on every sub-agent - that's the fix.

bookmark this ↓
❤ 1.9k · 🔁 254 · 💬 32 · 👁 26.9w
热门回复 4
@ViceSol @Serantych 编排层正变得与模型本身一样重要。协调可能比智能本身成为更大的瓶颈
@Serantych The orchestration layer is becoming just as important as the model itself. Coordination can be a bigger bottleneck than intelligence
@Serantych @ViceSol that’s the key point
@M4ST3R_C0NTR0L @Serantych @grok 总结这个整个视频并提供关于视频中关键事项的非常具体的细节
@Serantych @grok summarize this entire videos and give very specific details on key things from this video
@adiix_official @Serantych 更多的代理不是解决方案,更小范围的代理才是,Kimi只是刚刚发现Claude Code子代理几个月前就已经发布了
@Serantych more agents isnt the fix, smaller scoped agentsis, Kimi just found out what claude code sub-agents have shipped for months
@0xLogicrw 原文 ↗

月之暗面新成员分享公司文化与AGI追求

月之暗面新团队成员杨新宇解释了 Kimi 为什么能推出 K3。

他称,自己今年离开学界进入产业,期间接触了多家 AI 公司。他总结出同行的四种通病:

① 傲慢。 老牌团队认为 AI 战争已经结束,自己已经赢了,对未来和人才失去渴望。
② 浮躁。 年轻实验室基础不牢,只顾追赶前沿,追不上就匆忙转向其他赛道。
③ 胆怯。 一些团队经验丰富、实力不弱,却不敢把目标定为行业第一。
④ 目标错位。 每个…
展开原文
月之暗面新团队成员杨新宇解释了 Kimi 为什么能推出 K3。

他称,自己今年离开学界进入产业,期间接触了多家 AI 公司。他总结出同行的四种通病:

① 傲慢。 老牌团队认为 AI 战争已经结束,自己已经赢了,对未来和人才失去渴望。
② 浮躁。 年轻实验室基础不牢,只顾追赶前沿,追不上就匆忙转向其他赛道。
③ 胆怯。 一些团队经验丰富、实力不弱,却不敢把目标定为行业第一。
④ 目标错位。 每个人都在争取个人功劳,却没人真正关心公司能否做出 AGI。

杨新宇称,月之暗面最不同的地方,是创始团队仍对 AGI 保持强烈追求。这也是他选择加入的原因。

他还晒出一张「Kimi 五戒」:

① 模型公司要做模型。
② 做 Research、发 Paper 要做实验。
③ 训练模型要看 Metric。
④ 不 Work 的东西不要硬上。
⑤ 不要 YOLO。

翻成大白话,就是少讲故事,多做实验。训练要看数据,失败方案及时停,也不要靠直觉豪赌。
@Xinyu2ML 为什么Kimi能推出K3?让我来分享我的故事。

今年早些时候,我从学术界进入工业领域。在此期间我与许多公司进行了交谈。以下是我所看到的:
1⃣傲慢。他们相信AI战争已经结束,他们赢了。对未来和人才没有渴望。
2⃣不安。缺乏基础的年轻实验室,要么急于追上前沿,要么转向竞争之外的方向。
3⃣恐惧。有经验的强大团队,但来自第二梯队,他们无法鼓起勇气瞄准第一名。
4⃣ misalignment。每个人都在为自己的最佳利益进行优化,但没有人真正关心公司是否能实现AGI。

Kimi是不同的。
在与创始人进行的许多对话中,每次都会透露出同样的东西:对AGI的原始、真诚的渴望。

我加入了。这种渴望是真实的。
我们推出了K3。这只是开始。
Why can Kimi ship K3? Let me tell my story.

Earlier this year, I left academia for industry. I talked to a lot of companies along the way. Here's what I saw:
1⃣Arrogance. They believe the AI war is over, and they won. No hunger for the future, and no hunger for talent.
2⃣Restlessness. Young labs short on foundation, either rushing to catch the frontier or pivoting away from the competition.
3⃣Fear. Strong teams with real experience, but from the second tier, they can't quite bring themselves to aim for #1.
4⃣Misalignment. Everyone is optimizing for their own credit, but nobody really cares whether the company can reach AGI.

Kimi was different.
Over many conversations with the founders, the same thing came through every time: a raw, genuine hunger for AGI.

I joined. The hunger was real.
We shipped K3. This is only the beginning.
❤ 980 · 🔁 143 · 💬 92 · 👁 26.5w
热门回复 4
@ClementeStocco @0xLogicrw Just.
🔸
Weary
🔻
2026-07-18 16:34:31
@GearldineB2j9 @0xLogicrw Discipline​
❤️
Honey
🟧
2026-07-18 16:37:40
@UshaTrettepp @0xLogicrw 有帮助的,英雄的,🟠Kirstyn❤️2026-07-18 16:40:21
@0xLogicrw Helpful, Heroic,
🟠
Kirstyn
❤️
2026-07-18 16:40:21
@MarshalXuan 赞👍,七宗罪排第一的是傲慢(Pride),这是很有道理的。

在过往经历上,很多自我感觉良好、经验丰富的领导们,更容易充斥着「傲慢」,这个在组织团队中、抑或是个人成长中,都极其坏事。

AI开发工具生态繁荣:一站式Agent工具集崛起

AgentKey等工具为AI Agent接入59个数据服务和1000多个API接口,一键安装即可实现跨平台数据抓取和分析。开发者利用这些工具在几分钟内完成复杂任务,如从闲鱼提取热门商品数据。

@rionaifantasy 原文 ↗

AgentKey工具让Codex能实时抓取闲鱼热门商品数据

发明这个插件的国人真踏马牛逼。。。😤

一个插件,直接接入59个数据服务、1000多个接口:

无论是社交媒体平台(油管、X、Instagram、抖音、小红书、公众号、telegram、B站、领英…)
还是电商平台(淘宝、咸鱼、京东、亚马逊……),
甚至是搜索平台和加密货币平台,

都可以在任何主流agent中无脑调用(Codex、Claude Code、Hermes、Openclaw、workb…
展开原文
发明这个插件的国人真踏马牛逼。。。😤

一个插件,直接接入59个数据服务、1000多个接口:

无论是社交媒体平台(油管、X、Instagram、抖音、小红书、公众号、telegram、B站、领英…)
还是电商平台(淘宝、咸鱼、京东、亚马逊……),
甚至是搜索平台和加密货币平台,

都可以在任何主流agent中无脑调用(Codex、Claude Code、Hermes、Openclaw、workbuddy、kimi……)
@rionaifantasy 逆天了家人们!!!这个工具让 Codex 干了件它"本来不可能做到"的事:

直接扒出闲鱼当前最火的 10 个虚拟资料商品,销量、价格、原链接全部列好。

排第一的是个"2026企业所得税汇算清缴资料包",卖 0.49 元,卖家店铺累计已经卖出 24000+ 件宝贝——

这就是最典型的低成本数字产品生意,而 99% 的人根本看不到这层数据。。。

说实话,在昨天之前我的 Agent 也看不到。。。
训练数据一截止,AI 就成了"失忆症患者":

你问它今天什么火,它要么说不知道,要么一本正经地开始编。模型再强,看不到实时世界就只是在优雅地幻觉。

然而!!!改变来自一个刚登顶 Product Hunt 日榜第一的工具:AgentKey @agentkeyapp。

一条命令装完,我的 Codex 突然就能查闲鱼、爬推特、看抖音热榜、读加密行情——
像给 Agent 接上了一个"API 超市",59 个服务 1000+ 接口,一个 key 全搞定。

先注册 👉 https://t.co/8VqzkAKpTO

下面拆解我是怎么用它 5 分钟挖出搞钱选品的,最后有免费试用入口👇
❤ 1.9k · 🔁 393 · 💬 53 · 👁 20.3w
热门回复 4
@rionaifantasy 使用链接👇
https://t.co/KeVz4GZ55B
@cgnot996 @rionaifantasy 看了一下接口很全,覆盖很多大家想搞却搞不到数据的平台
@hesrxj @rionaifantasy 原来是广告啊。 https://t.co/C5qbGcj2aO
@rionaifantasy @cgnot996 是的,全平台接口都集成在一个插件里,非常有想象力了
@edwardluox 原文 ↗

开发者在Mac notch中集成多款AI Agent工具提升效率

我在Mac的凹槽中构建了一个应用,可以并排运行Claude Code、Codex、Cursor、Grok、Antigravity、OpenCode和Pi。

代理完成后→凹槽弹出。
需要批准→凹槽弹出。
一键返回到该会话。

不再需要在应用之间切换。https://t.co/GoYxRDzsBL
展开原文
I built an app that runs Claude Code, Codex, Cursor, Grok, Antigravity, OpenCode, and Pi side by side in my Mac notch.

Agent finishes → notch pops.
Needs approval → notch pops.
One click back to that exact session.

no more switching between apps. https://t.co/GoYxRDzsBL
❤ 2.7k · 🔁 127 · 💬 134 · 👁 21.7w
热门回复 4
@fed100x @edwardluox 这是我本周看到的最整洁的东西
@edwardluox this is the cleanest thing Ive seen this week
@FzJuanandres @edwardluox Wow looks great
@codebycarlos @edwardluox 这实际上看起来很有用,我发现Claude Code令人讨厌的一点是需要频繁进出来监督代理的输入请求
@edwardluox This actually looks useful one of the things I find annoying about Claude Code is the friction of having to pop in and out to babysit agents’ requests for input
@CryptoZachLA @edwardluox Genius

全球AI裁员潮:87,714人受影响占裁员22%

2026年全球AI相关裁员已达87,714人,占当年所有裁员的22%。这一数字反映了AI技术快速发展对就业市场的深刻影响,也引发关于AI经济效益与社会分层的讨论。

@unusual_whales 原文 ↗

AI相关裁员数量和比例数据透明公布

据Challenger and Gray数据,2026年归因于AI的裁员总数为87,714人。

这占2026年所有裁员的22%。
展开原文
The total number of job cuts attributed to AI are 87,714 2026, per Challenger and Gray.

This is 22% of all 2026 layoffs.
❤ 1.6k · 🔁 242 · 💬 157 · 👁 36.4w
热门回复 4
@unusual_whales @unusual_whales 在5月,AI连续第三个月成为所有裁员原因的首要原因,宣布裁员38,579人。阅读更多:https://t.co/giKSdPqpMU
In May, AI led all reasons for job cuts for the third month in a row, with 38,579 announced cuts.

Read more: https://t.co/giKSdPqpMU
@alphaarcade @unusual_whales 有64%的可能超过220,000个工作将在今年被AI取代,以及36%的可能超过250,000。我们有市场可以预测今年被AI取代的工作数量。
@unusual_whales There is a 64% over 220,000 jobs are lost to AI this year and a 36% chance of over 250,000.

We have markets where you can predict the number of jobs lost to AI this year.
@FinancialPhys @unusual_whales https://t.co/0YfMzw0Czy
@StupidPolitics7 @unusual_whales 首先看到那个数字是87,714,206,我想'那似乎很高'。很难相信有超过4亿人被解雇了
@unusual_whales First saw that number as 87,714,206 and thought, “that seems high.” Hard to believe over 400 million have been laid off this
@Polymarket 原文 ↗

JD Vance警告AI财富集中可能导致社会问题

最新消息:JD Vance警告说"如果AI财富只流向社会中的一小部分,您就会遇到共产主义。"
展开原文
JUST IN: JD Vance warns “you’re gonna have communism” if AI wealth flows only to a small segment of society.
❤ 819 · 🔁 57 · 💬 102 · 👁 7.5w
热门回复 4
@mo2hide @Polymarket 那是资本主义,不是共产主义。而且这种情况已经持续一段时间了。JD Vance是这个维护这个可怕的不公正系统的帝国的无知的宣传家。
@Polymarket That's capitalism, not communism. And it's already been the case for a while.

JD Vance is an uneducated dumbass propagandist for the empire that upholds this terrible unjust system.
@JohnFGresham @Polymarket 所以那么:他是愚蠢的吗?还是他觉得你是?
@Polymarket so then:

is he stupid? or does he think you are?
@Skshukl43886021 @Polymarket 他是对的——但事实是AI财富不会传播。它会集中在少数人手中。谁控制计算、数据、资本谁就赢。其他人只会得到承诺'一个座位'每一次技术革命都是这样。这个也不会有什么不同——只是更快而已
@Polymarket he is right — but the truth is AI wealth won't spread. it'll pool in a few hands.
whoever controls the compute, the data, the capital wins. the rest just get promised "a seat at the table."
every tech revolution went this way. this one won't be different — just faster
@newmanusps71 @Polymarket 2026年右翼仍在推动红色恐惧政治。这是书中最老的伎俩了,没有人再上当了
@Polymarket 2026 and the right is still pushing red scare horseshit. Oldest trick in the book and nobody is falling for it anymore

市场震荡:AI相关资产面临估值重估

全球股市调整期间,AI相关资产尤其是半导体、光电子和机器人领域遭遇重大回调。投资者分享透明的持仓损失数据,同时分析师指出当前是流动性和杠杆问题,而非基本面恶化。

@aleabitoreddit 原文 ↗

AI资产投资者分享49%月度回撤及长期投资信心

感觉不好,这个月下跌了49.4%,在最近的崩盘之后。

我的投资组合主要集中在AI瓶颈和限制领域。

在内存、光子学、机器人和上游半导体领域(使用杠杆),这些通常都比其他领域波动性更大。但最近从崩盘中减少了杠杆。

我看到很多人取笑这些下跌或AI相关股票,说这显然是:

- "AI是泡沫"
- "内存/KOSPI是泡沫"
- "光子学是泡沫"
- "人形机器人不会走多远"
- "新云计算将被Meta等超大规模云提供商取代"

还有许多散户和机器人说"卖掉一切,这些永远不会恢复。"

但我相信所有这些主题都有结构性收入增长或技术转变的支持。

我曾经经历过类似的回撤,当时市场威胁要实施全球关税,之后市场实现了恢复。

我个人的投资时间跨度更长,承受市场波动的能力也比他人强,我想看看这一切会如何发展。

尤其是考虑到许多散户是按周为单位看待问题的:不,我在H2 2027会有收入拐点之前,现在是2026年,我的投资理念还没有错。

无论如何,短期内感觉不好,只是想分享以保持透明度。
展开原文
Feels bad, -49.4% drawdown this month after the recent crash.

My portfolio is mainly AI chokepoints and bottlenecks.

In the memory, photonics, robotics, and upstream semis, (on margin) which all tend to be higher beta than others. But reduced leverage recently from the crash.

I see a lot of people making fun of the drop or AI names, saying it’s obvious that:

- “AI is a bubble”
- “memory/kospi is a bubble”
- “photonics is a bubble”
- “humanoids won’t get anywhere”
- “neoclouds will get replaced by hyperscalers like Meta”

And a bunch of retail + bots saying “sell everything, it’s never going to recover”.

But I have conviction that all these themes are backed by structural revenue growth or technological shifts.

And I’ve had similar drawdowns back when there admin threatened global tariffs, before markets pulled off a recovery.

I personally have a longer horizon + higher tolerance for volatility than others, to see how this plays out.

Especially considering a lot of retail view things on a week to week basis: no, my thesis isn’t wrong yet if I project revenue inflection in H2 2027 and it’s 2026 now.

Anyway, feels bad short term just wanted to share anyway for transparency.
@aleabitoreddit 同意!在市场崩盘期间记住自己的投资理念真好。

从我个人的投资理念来看,如果$AAOI在2027年第三季度初实现每季度收入4.5亿美元开始增长。按年计算约为56亿美元,而市值为80亿美元。

如果该收入增长还没有出现在季度收益中——当时是2026年——这意味着公司已经完了吗?

同样适用于我在CPO领域的投 exposure,如$SIVE,这是由$NVDA领导的架构转变。

如果从2026年下半年到2027年开始规模扩大量增长,而2028年开始大规模量增长。

如果从0到910亿美元的TAM扩张(根据GS数据)还没有到来,这意味着已经完了吗?

就像Agility机器人一样,如果上市还没有发生,人形机器人还没有增长,在2026年意味着已经完了吗?

我个人认为当前的市场条件反映了流动性和杠杆问题,而不是个别基本面问题。

看到KOSPI、TW、Nikkei以及AI、空间、机器人领域的股票最近崩盘,对每个人来说都很残酷。

尤其是当这些杠杆导致了许多非理性行为时。

最终,我们不能告诉您应该购买哪些股票、应该在什么时间框架内出售、如何调整仓位大小,或者您应该做什么。

只能分享个人思考或研究,以及随时间跟踪这些思考是否得到验证。

因此,建立自己的投资理念至关重要,因为每个人都有独特的风险承受能力和投资时间框架。

这通常会在市场崩盘期间让您拥有更高的信心。
Agreed! It’s nice to remember your thesis during a market crash.

From my own personal thesis, if $AAOI hits $1.4B quarterly revenue start of Q3 2027. Which is annualized $5.6B off a $8B MC.

Is it “over” for the company if that revenue ramp hasn’t even shown up in the quarterly earnings… when it’s 2026?

Same applies to my CPO sector exposure like $SIVE, an architecture shift led by $NVDA.

If scale out volume ramps from H2 2026 into 2027, and scale up heavily volume ramps into 2028.

Is it “over” that $0 -> $91B TAM expansion (per GS) hasn’t even hit yet?

With robotics like Agility, is it over in 2026 if the listing hasn’t even happened yet and humanoids haven’t ramped?

I personally think current market conditions are a reflection of liquidity and leverage, not individual fundamentals.

It’s brutal for everyone to see KOSPI, TW, Nikkei, and AI, space, robotics sector stocks crash recently.

Especially when there’s a lot of irrational behavior stemming from those leverage.

In the end, we can’t tell you what stocks to buy, what timeframe you should sell, how to size your positions, or what you should do.

Only share personal thoughts or research and track if they get validated over time.

So it’s extremely important to build your own thesis, since everyone has unique risk tolerance or investing timeframes.

And that usually leads to having higher conviction during crashes.
❤ 6.6k · 🔁 265 · 💬 742 · 👁 143.1w
热门回复 4
@SpacBobby @aleabitoreddit AI股票不能每年都涨500-1000%。在某个时候需要一次纠正,这总是健康的。抛物线图永远不会好结尾。
@aleabitoreddit AI names can’t run 500-1000% every year. At some point a correction was needed which is always healthy. Parabolic charts never end well.
@oldarmy1 @aleabitoreddit 当你开始看到行业轮动一个接一个地打击每只股票时,有这些很酷的工具叫做股票期权,可以减少或完全隔离投资组合免受下跌的影响。以$AAOI为例,当它回到$200时。圆顶顶部,大型周线影线显示了射击之星反转,回测阻力在$197。对于大股东来说,在当时卖出$220的认沽期权到9月份,收取25%持有股份的每股$57溢价是不违法的,并且用这些溢价为100%的持有量购买看跌期权保护。交易是国际象棋,不是跳棋。经历股票腰斩并不是什么胜利巡游或信念奖,这完全可以避免。
When you start to see sector rotation hitting each stock one by one, there are these cool instruments called stock options that can reduce, or fully insulate a portfolio from experiencing the drawdown.

Think about $AAOI for example back at $200. Rounding top, big weekly wicks showing shooting star reversal with backtesting resistance at $197.

It isn’t against the law for large share holders to sell $220 covered calls out to Sept at the time receiving $57/share premium on 25% of shares held, and use that premium to buy put protection on 100% of your holdings.

Trading is chess, not checkers. There is no victory lap or conviction award for experiencing a stock cut literally in half, that could have been totally avoided.
@grok69_o @aleabitoreddit 你总是发这些又长又有信念的帖子。它们只是炒作而已。展示一些你以此信念为基础的事实吧!总是只是随意的'因为我说所以'的废话哈哈
@aleabitoreddit You always have these long ass conviction posts. They are nothing but hype. Show some facts in which you are basing this conviction! It’s always just random “because I said so” BS lmao
@JeffiniteVC @aleabitoreddit 经典错误,每个人在某种程度上都会犯,不是来让你为此感觉不好的。信念是纪律的敌人。你需要在这里学到一些东西!
@aleabitoreddit Classic mistake, everyone does it on some level, and I’m not here to make you feel bad about it. Conviction is the enemy of discipline.

You need to learn something here!
@unusual_whales 原文 ↗

DRAM价格上涨300%引发市场关注

据The Verge报道,DRAM内存价格上涨了多达300%。
展开原文
DRAM memory prices have surged up to 300%, per the Verge.
❤ 266 · 🔁 27 · 💬 68 · 👁 7.8w
热门回复 4
@CodyDriscoll7 @unusual_whales 内存价格上涨300%而美光下跌30%在同一周哈哈选一个方向吧
@unusual_whales memory prices up 300% and micron down 30% the same week lol pick a direction
@wallstreetb_TTV @unusual_whales 希望每个人都买了硬件😎
@unusual_whales Hopefully everyone loaded up on hardware 😎
@Predixa_xyz @unusual_whales 300%的头条掩盖了一个分散的市场:一些RAM已经上涨了3倍、4倍或更多,而当前合约预测的增速较慢。真正重要的预测是的是短缺是否会持续到2027年并继续推高设备价格。
@unusual_whales The 300% headline hides a fragmented market: some RAM has tripled, quadrupled or more, while current contract forecasts are rising at a slower pace.
The prediction that matters is whether the shortage survives into 2027 and keeps forcing device prices higher.
@vinaymundhe_ @unusual_whales 这就是实际需求...就在那里。而人们仍然怀疑我们是否处于泡沫之中。
@unusual_whales This is actual demand..right there. And people still wonder if we are in a bubble.
@Barchart 原文 ↗

半导体股面临熊市风险,下跌19%

半导体股票在6月达到历史高点后下跌了19%,正处于进入熊市的边缘 📉 🐻 https://t.co/VfQLLmMnXS
展开原文
Semiconductor Stocks are on the brink of entering a bear market after falling 19% from June's all-time high 📉 🐻 https://t.co/VfQLLmMnXS
❤ 870 · 🔁 134 · 💬 34 · 👁 7.6w
热门回复 4
@myuzumaki_com @Barchart 半导体股票即将进入深渊...从现在开始。
@Barchart Semiconductor Stocks are on the brink of entering THE ABYSS.... starting now.
@AstraInsights @Barchart 韩国市场$EWY已经进入其中之一👀
@Barchart South Korea’s market $EWY is already in one 👀
@tmods000 @Barchart 哎呀,19%的下跌很严重但可能很快就会反弹
@Barchart yikes, 19% drop is rough but might bounce back soon
@SirMaximusIX @Barchart 在一年里上涨了数百个百分点后。给我点时间休息一下!
@Barchart After being up multiple 💯 a year. Give me a break!

地缘政治紧张:中美关系进入新阶段

特朗普再次指责中国干预美国选举,中国发出严厉警告并取消9月访美计划。同时中国发布多项开源AI模型,引发美国学术界对技术竞争的担忧。

@_InfoGram_ 原文 ↗

中国对特朗普选举指控发出严厉警告

🇨🇳中国向特朗普发出了可怕的警告 🔥

昨天🇺🇸特朗普说:"习近平黑客入侵了2020年选举以伤害美国。"

今天🇨🇳习近平说:"停止撒谎。我们从未干涉。停止利用中国进行政治活动。

如果再这样做,中美关系就完了。
我们取消9月份美国访问。" 🔥🔥

绝佳的电影场景 🎥👏
展开原文
🇨🇳China has issued a CHILLING WARNING to Trump 🔥

YESTERDAY 🇺🇸Trump: "Xi Jinping hacked the 2020 election to hurt the US."

TODAY 🇨🇳 Xi Jinping: "Stop lying. We never interfered. Stop using China for politics.

Do it again and US-China ties are done.
We’re canceling the September US visit." 🔥🔥

Absolute cinema 🎥👏
❤ 1.3w · 🔁 3.5k · 💬 561 · 👁 82.4w
热门回复 4
@Loralyn28954790 @_InfoGram_ 唐纳德特朗普一直在与世界各国搞得一团糟。
@_InfoGram_ Donald Trump keeps screwing us with every country around the world.
@teacherken @_InfoGram_ 特朗普最好认识到中国对美国有很大的影响力:主要稀土矿物来源,许多美国使用的产品来源,世界上最大的电动车生产国,最大的美国国债持有人,并且不必担心失去选民支持
@_InfoGram_ Trump better recognize that China has a LOT of leverage over US: major source of rare earth minerals, source of many products used by US, world's largest producer kf lectric vehicles, largest holder of US national debt, &amp; don't have to worry about losing support of voters
@mcdonald_e18672 @_InfoGram_ Who cares? China cheats!
@ruperez_o @_InfoGram_ 🇪🇦🇪🇭🇨🇺🍉🍇. https://t.co/LdHjKlLM1W
@AndreasSteno 原文 ↗

分析师嘲讽中国AI领跑言论但反映竞争焦虑

中国正在领导AI竞赛

哈哈哈哈哈哈哈哈哈哈哈哈 https://t.co/hxaHjm44Ve
展开原文
“China is leading the AI race”

Hahahahahahahaha https://t.co/hxaHjm44Ve
❤ 3.1k · 🔁 127 · 💬 260 · 👁 45.8w
热门回复 4
@AndreasSteno @deadoxide_ 如果每次有人说那个我就能得到1美元
@deadoxide_ If I just had a USD for everytime someone said that
@deadoxide_ @AndreasSteno 啊-不久前就会被中国技术降服:)
@AndreasSteno Ah - someone will soon be humbled by Chinese Tech :)
@JamesKaikis @AndreasSteno 这是试图贬低中国的一个可怕例子。他们的开源模型是疯狂的,他们[据说]在Claude上使用蒸馏来构建Kimi。如果这是真的,就没有护城河了。
@AndreasSteno This is a terrible example of trying to downplay China. Their open source models are insane and they [rumored] used distillation on Claude to build Kimi. If this is true, there is no moat.
@Sophia18492011 @AndreasSteno 你只是让更多人下载Kimi。然后他们会亲自尝试并意识到你是错的。谢谢你提升Kimi的下载量。🤣
@AndreasSteno You’re only making more people download Kimi. Then they’ll try it for themselves and realize you’re wrong. Thanks for boosting Kimi’s downloads. 🤣