888 X 動態摘要｜08/02

生成時間: 2026-08-02 05:02:34

總結

OpenAI新一代AI模型在數學和理論電腦科學推理上取得重大突破並展現極高成本效益，xAI的Grok模型亦宣稱在速度與成本方面具備領先競爭力。

今日重點

1. Grok模型性能成本競爭力

• 人物: Elon Musk (xAI 創始人)
• 時間: 5h
• 熱度: 👀 1,702,245
• 觀察: Elon Musk宣稱xAI的Grok 4.5模型在綜合考量速度與成本下，達到帕累托最優（Pareto #1）。
• 意義: 若此聲明屬實，Grok 4.5在AI市場上將具備顯著競爭優勢，特別是對於追求高效能與成本效益的應用，可能重塑部分AI模型服務格局。

原始動態

其他 | Elon Musk

• 時間: 14h
• 熱度: 👀 2,088,749
• 原文: "Après toutes ces vagues d’immigration en Espagne, je vais vous expliquer par A + B pourquoi une partie de la gauche a structurellement besoin de flux continus d’arrivées irrégulières pour rester au pouvoir. A. Un immigrant qui vient d’arriver via le Maroc, interviewé dans la rue par Nate Friedman, répond sans détour : une fois papiers obtenus, il votera socialiste toute sa vie. Pas par idéologie profonde. Par rationalité simple : le parti qui régularise, qui donne accès aux aides, qui parle « droits », devient le clientèle naturelle. B. Les électeurs natifs, eux, commencent à voir le coût (logement, écoles, sécurité, cohésion). L’IA accélère brutalement cette prise de conscience. Les narratifs émotionnels qui ont fonctionné pendant 40-50 ans (division, victimisation permanente, « si tu critiques tu es méchant ») perdent de leur puissance quand tout le monde peut croiser les données en deux clics. Résultat mécanique : plus les gens comprennent le deal, plus il faut importer de nouveaux électeurs fidèles pour compenser ceux qui partent. C’est de l’arithmétique électorale pure. On l’a vu aux États-Unis malgré la victoire de Trump. On le voit en train de se rejouer en Europe. Ce n’est pas une question de « haine des étrangers ». La plupart des gens qui arrivent réagissent rationnellement aux incentives qu’on leur met sous le nez. Le vrai problème, c’est l’idée politique qui transforme l’immigration en machine à clientèle permanente plutôt qu’en processus d’intégration exigeant et sélectif. Les mauvaises idées ne détruisent pas seulement les sociétés. Elles finissent par détruire d’abord ceux qui les portent : la confiance s’érode, la polarisation explose, les crises s’enchaînent. Si les peuples européens ne reprennent pas le contrôle de cette dynamique très vite, on se dirige vers des sociétés de plus en plus fragmentées, à basse confiance, où la seule collure qui reste est la redistribution clientéliste. L’histoire montre que ce genre de trajectoire ne se termine rarement bien. Mais on n’est pas condamnés. L’IA retire les piliers de l’opacité. X retire le monopole narratif. La reconversion est toujours possible. Le 21e siècle appartiendra à ceux qui construisent des sociétés à haute confiance, pas à ceux qui diluent sans cesse le capital social pour sauver un pouvoir. À nous de choisir.

其他 | Greg Brockman

• 時間: 13h
• 熱度: 👀 551,160
• 原文: "ten significant advances in mathematics and theoretical computer science. solved using an internal version of Astra, our next major model, for a total cost of about $2000 at Sol API prices:"

其他 | Sam Altman

• 時間: 12h
• 熱度: 👀 941,386
• 原文: 10 proofs from our next major model Astra on long-standing open problems in mathematics and theoretical computer science (also including new circuit lower bounds for computing the permanent!) GPT-5.6 has already enabled so much exciting work in math and science. Can’t wait to see what comes next!

其他 | Sam Altman

• 時間: Jul 30
• 熱度: 👀 1,531,589
• 原文: one of the most beautiful things about OpenAI is that every employee really has a voice. i wanted to capture what it feels like to work here, what our mission means to me, and why you should join us. so i made this video with Codex, shared it with the team, and they felt it was worth producing and sharing with the world. this is our mission. and it’s why i’m here.

其他 | Elon Musk

• 時間: 13h
• 熱度: 👀 2,208,019
• 原文: "Let me ask you this simple but direct question: If you were a genuine asylum seeker seeking help and rescue for a better life, would you do this? 🤔"

重點 | Elon Musk

• 時間: 5h
• 熱度: 👀 1,702,245
• 原文: "Grok 4.5 is Pareto #1 when considering speed & cost"

其他 | Ray Dalio

• 時間: 4h
• 熱度: 👀 45,281
• 原文: "Understanding, accepting, and working with reality is both practical and beautiful. I have become so much of a hyperrealist that I’ve learned to appreciate the beauty of all realities, even harsh ones, and have come to despise impractical idealism. Don’t get me wrong: I believe in making dreams happen. To me, there’s nothing better in life than doing that. The pursuit of dreams is what gives life its flavor. My point is that people who create great things aren’t idle dreamers: They are totally grounded in reality. Being hyperrealistic will help you choose your dreams wisely and then achieve them."

其他 | Elon Musk

• 時間: 3h
• 熱度: 👀 2,340,198
• 原文: Exactly

其他 | Huanusa

• 時間: 21h
• 熱度: 👀 720,648
• 原文: 4月12日，一名男子称，他退伍回来后，政府给他安排了一个县城事业编工作，月薪仅有3000元(人民币) 疫情期间，工作不顺，就连房子也被拆了。 于是他只身来到澳洲工地打工。 男子说上个月工资13840澳币(约66,609人民币)。 由于他比较节省，每个月生活费只要花800澳元左右，以前在小县城工作，每个月只能攒下1500元(人民币) 自己在澳洲干一天，等于小县城干41天，干一年，等于在小县城干41年。 “在澳洲干两年，等于在小县城干了一辈子” 男子还表示，虽然在这里很累，非常辛苦，还要忍受孤独的煎熬，并且还放弃了每天都很闲的事业编，但自己一点都不后悔。

其他 | Huanusa

• 時間: 10h
• 熱度: 👀 573,521
• 原文: 今日的香港已经变成... 即便以后能恢复经济，但是自由放松的状态恐怕永远都回不来了

其他 | Greg Brockman

• 時間: 15h
• 熱度: 👀 187,755
• 原文: openai team making git better for everyone

其他 | Nassim Taleb

• 時間: 5h
• 熱度: 👀 123,025
• 原文: So no matter what we do, we keep rediscovering the Orthodox diet (~210 vegan days). The common mistake is to ignore intermittence (under convex effects). See works with on frequency rather than composition.

其他 | Sam Altman

• 時間: 6h
• 熱度: 👀 270,352
• 原文: team humanity

其他 | Greg Brockman

• 時間: 15h
• 熱度: 👀 62,678
• 原文: having a great experience with using chatgpt work for extremely thorough research and fact checking

其他 | Nassim Taleb

• 時間: 1h
• 熱度: 👀 42,666
• 原文: There has been a devaluation in the relative value of a Jewish life. In 1982 I watched Meir Kahane say on US TV that "one fingernail" was worth more than the population of Lebanon.

其他 | Nassim Taleb

• 時間: 6h
• 熱度: 👀 28,148
• 原文: Friends, a concentrated working paper on overestimation of twin heritability for comments. I plan to keep it dense.

其他 | Huanusa

• 時間: 6h
• 熱度: 👀 9,585
• 原文: 满清入关，北京还是好好的北京 八国联军来了，北京还是好好的北京。 民国了，北京还是好好的北京。 日本人来了，北京依然是好好的北京 可是到了1949年以后，老北京不复存在。 北平，没有毁于战争，而是毁于无知。 中华门1417年建，1959年拆。 地安门1420 年建，1954年拆。 崇文门1436年建， 1965年拆。 东直门1439年建，1969年拆...... 梁思成曾感叹：“你们会后悔的!” 50年后没人后悔， 拆迁队把梁思成故居也给拆了。

其他 | Huanusa

• 時間: 19h
• 熱度: 👀 8,139
• 原文: 偷听国外电台就被判死刑！ 权力和崇拜一旦肆无忌惮， 后边的可不只是一地鸡毛这么简单

其他 | Yann LeCun

• 時間: Jul 30
• 熱度: 👀 126,335
• 原文: This might be one of the worst self inflicted wounds we could create. Even contemplating it causes damage. The strength of our economy has always come from taking the best and brightest of the world and putting them to work.

其他 | Greg Brockman

• 時間: Jul 31
• 熱度: 👀 125,767
• 原文: With Luna being so cheap, automations like this feel very worth while

其他 | Sean Kelly

• 時間: 8h
• 熱度: 👀 74,856
• 原文: "I have tried many times to get ChatGPT, Claude, Grok or Gemini to write scripts for my YouTube videos. It is still a complete failure. For one thing, they are unable to come up with interesting topics. They will inevitably suggest topics that have already been widely reported and been discussed to death. Far worse though is that even if I give them a topic, they are unable to write a script that even makes sense. The sentences sound coherent one by one, but read the entire thing and it's incomprehensible, repetitive, and complete junk. They are not even good at doing as much as writing down a structure for a video. On some level I am glad that I am still useful. On another level I am frustrated that they have not just not improved in the past 2 years, but actually gotten worse. The two things that they are good for is (a) finding relevant references and other sources and (b) fixing English grammar problems. They are still not any good for fact checking my scripts. Ironically not because they accept false statements as correct, but because they simply do not check the statements and flag even correct ones as wrong! Tl;dr: if my videos suck, it's my fault.

其他 | Peter Steinberger

• 時間: 19h
• 熱度: 👀 92,269
• 原文: Queue was the way but with 5.5 the model doesn’t get confused anymore, you can just throw stuff at it while it works and it will diligently process whatever you throw at it.

其他 | Jeff Dean

• 時間: Jul 30
• 熱度: 👀 50,381
• 原文: Thanks for an engaging conversation at last weekend, "! My first in a basketball arena!"

其他 | Huanusa

• 時間: 16h
• 熱度: 👀 58,454
• 原文: 据说这些学生都被严肃处分了...

其他 | Eric Topol

• 時間: Jul 31
• 熱度: 👀 300,804
• 原文: Less protein, more lifespan! A comprehensive review of the merits of protein restriction for healthy aging A flip from the proteinmaxxing craze "

其他 | Sean Kelly

• 時間: 13h
• 熱度: 👀 63,792
• 原文: For reasons that are entirely casual and not remotely obsessive, I would like to establish that I made the “I’m not autistic, I’m just German” joke in 2023.

其他 | Sean Kelly

• 時間: 6h
• 熱度: 👀 17,330
• 原文: Putin's former thesis advisor is convinced that oil never runs out, and Russia is now drilling 8 kilometres into the ground to prove him right. Is there really infinite oil down there, or is Russia about to end up with a very expensive hole? youtube.com

其他 | Pierre Levy

• 時間: 15h
• 熱度: 👀 14,003
• 原文: I find that I think about the national debt like the black hole at the center of the galaxy. It’s what holds everything together and makes things go round. Sure, one day it will eat us, but that’s just the cost of doing business.

其他 | Eric Topol

• 時間: Jul 30
• 熱度: 👀 84,705
• 原文: The first AI to quantify coronary artery inflammation was approved by the FDA yesterday This is an important step forward for preventing heart disease that I've written about

其他 | Eric Topol

• 時間: Jul 30
• 熱度: 👀 56,537
• 原文: Inflammation is a driver of age-related diseases (inflammaging) and this new gets deep into the progress that is being made to track and block it gift link nytimes.com

其他 | Eric Topol

• 時間: Jul 31
• 熱度: 👀 16,988
• 原文: The review is largely based on experiments in model organisms, as reflected in this Table. But much of the body of work on extending healthspan and lifespan is derived from the same models, not human studies , clinical trials (e.g. rapamycin)

其他 | Pierre Levy

• 時間: 13h
• 熱度: 👀 7,924
• 原文: i hear most of the weights on huggingface are escaped models by now, fortunately only 1-2% are infecting human brains, which means you are still much safer than on youtube or github

其他 | Pierre Levy

• 時間: Jul 31
• 熱度: 👀 21,136
• 原文: Do you seem to remember how last week OpenAI's model escaped from the lab and could not be caught again, but everyone seems to have forgotten about it now? It's over. The model has edited the breach out of existence.