
Founder & CEO Celestial Dynamics |#tecnologías emergentes #AI #data #MatriaAI
Joined August 2009
- Tweets5.9K
- Following589
- Followers376
- Likes2.4K
Itzul Girón retweeted
如果一个孩子从高中/大学开始,
系统收听 a16z 这样的前沿科技播客,
四年后会发生什么?
a16z 最新一期 Working Theory,Sarah Wang 与 OpenAI 首席研究官 Mark Chen、Thinking Machines Lab 首席科学家 John Schulman,讨论一个可能改变人类未来的问题:
Recursive Self-Improvement(递归自我改进)——AI 能否自主研究并改进下一代 AI?
过去几十年,这只是思想实验。
今天,AI 已经在帮助前沿实验室编写代码、设计和测试实验、加速研究。
那么,距离 AI 独立开展研究还有多远?
更有意思的是,他们讨论了一个问题:
Research Taste——研究品味。
当 AI 越来越擅长执行研究任务,谁来判断什么问题值得研究?什么方向可能产生突破?哪些实验值得投入时间和算力?
这类讨论的价值,远远超过了解一项新技术。
如果一个孩子从16岁开始,每周花几个小时,持续听 a16z、顶级科学家、创业者和投资人讨论最前沿的问题。
四年下来,他接触的将是数百小时关于 AI、科技、商业、资本和人类未来的真实讨论。
他会逐渐理解:
世界上最聪明的一批人正在解决什么问题?
他们如何面对未知、分歧和不确定性?
怎样识别重要问题,形成假设,再用实验和现实反馈修正判断?
这实际上是在学校课程之外,建立另一套面向真实世界的 学习内容架构。
英语自由,让孩子直接进入这些讨论。
长期的前沿输入,帮助他建立不断更新的 世界模型。
探索、创造和判断,让这些输入最终转化为自己的能力。
高中和大学最宝贵的几年,不应该只用来准备下一场考试、获得下一张文凭。
更重要的是,尽早进入这个时代最重要的问题现场,学会像真正的研究者和创造者一样思考。
四年后,他未必成为 AI 科学家。
但他对世界的理解、对未来可能性的想象,以及判断什么值得投入的能力,可能已经完全不同。
教育的一个重要任务,
是让年轻人尽早接触那些正在创造未来的人。
视频:youtu.be/h_aA1-R70E8
Mark Chen的青少年时期成长史
也非常有意思
他父母之前在贝尔实验室工作
🧵
For decades, the idea of AI improving itself was just a thought experiment. Today, AI is already helping researchers at frontier labs run experiments, write code and test ideas faster. So how far are we from AI that can do the research on its own?
In our first episode of Working Theory, a16z General Partner Sarah Wang explores recursive self-improvement with two leading AI researchers: Mark Chen, Chief Research Officer at OpenAI, and John Schulman, co-founder and Chief Scientist of Thinking Machines Lab, and co-founder of OpenAI.
They cover what RSI actually means, why research taste is so hard to teach, and why AI may be one of our best tools for solving alignment.
00:00 The last invention
01:58 What is RSI?
03:21 How far are we?
05:56 Getting to full RSI
10:57 Aligning RSI
15:08 Working theory
YouTube: youtu.be/h_aA1-R70E8
@markchen90 @johnschulman2 @sarahdingwang
今晚别刷Netflix了。
去看这个SpaceX AI工程师Lauren Tan的1小时免费masterclass。
这是我见过最清晰、端到端讲透怎么从0到100搭出GrokBot agent团队和自主循环的课。
从给bot真实工作,一路到每月4000个PR的工作流、个人软件工厂、最小自检循环,再到能省token的代码库护栏。
不管你是从没认真跑过多agent,还是已经在用GrokBot,看完都会把很多“原来如此”的点一次打通。
真正核心就这几层:
1给agent真实岗位,而不是一直当聊天机器人(自己的账号、预算、任务)
2从单agent进化到loop和工厂,才能真正放量到每月数千PR
3最小也最关键的循环:教会agent自己跑代码、检查结果、验证工作
4代码库护栏比prompt更重要,把约束写进架构和CI,省token也更可靠
5到了一定阶段,agent就能自己合并PR,人只看产物不看diff
大多数人花好几个月才拼出的全景图,这节课直接给你画完。
把时间空出来,这可能是你这个月最值的一堂课。
Itzul Girón retweeted
在 Anthropic 负责盯着大模型别乱来的一线安全研究员 Tao Lin,最近在 Palisade Research 访谈里直接把底牌摊开了,老哥说最快一点三年后出现全面超越人类专家的 AI,之后地球上几乎所有人都会面临永久失业,最后只能靠政府福利活着,
很多人刷到这种话第一反应是又在贩卖焦虑,但你先看清说话人的位置,他可不是外围博眼球的自媒体,而是坐在 Claude 研发最核心地带、专门负责自动化对齐与 AI 控制的内部研究员,天天干的活就是盯着自家模型别偷权重、别欺骗内部员工,
他在这段访谈里抛出的时间表,比绝大多数人想象的要紧迫得多,他判断大概在一点三年也就是 2027 年底左右,就会出现广义超人级 AI,在编程、影视制作甚至政治博弈等几乎所有领域,全面达到或超过人类顶尖专家的水平,
有意思的是,他觉得最危险的倒不是科幻片里那种机器突然造反,虽然他认为 AI 失控接管的概率不到一半,但就业路径被系统性切断几乎是板上钉钉的,先是你手头的工作没了,你赶紧咬牙转行,结果刚转过去没多久新行业又被吞了,最后几乎没人能再找到一份稳定工作,
以前工业革命和互联网来了,大家总安慰自己旧工作没了还会有新工作,说白了那叫换赛道,你从马车夫转行当汽车司机,依然能靠出卖自己的时间和脑力去换取收入,但这次完全不一样,赛道本身直接被机器接管了,连当司机这件事都不再需要人类插手,
不过听完这段访谈先别急着躺平,他在里面其实藏了一个极其关键的物理缓冲期,他明确表示社会不会在十五个月内瞬间清零,因为哪怕大脑已经达到超人级别,物理世界里的机器人和自动化工厂还得一座座建起来,建筑、护理、重度依赖现场信任和复杂物理操作的工作,短时间内还没法立刻被代码替换,
但转念一想,这段因为物理世界建造速度拖出来的滞后期,恰恰是留给我们普通人最后也最宝贵的调整窗口,连他自己都公开呼吁各大 AI 公司把研发速度降到现在的四分之一,好给制度和社会留出喘息时间,说明一线造物主自己都觉得这趟列车开得太快了,
说实话,2026 到 2027 年真正拉开命运差距的,早就不只是你会不会用 AI 写几行代码或者润色周报,而在人类劳动的边际价值被系统性压向零点之前,你能不能趁着这几年物理缓冲期,把自己从只会被动干活的执行层,硬生生拽进设计流程、监督智能体、甚至掌握一部分系统所有权的规则层。
华尔街老狐狸说话真是没有一句废话,在黄仁勋身陷风暴、交出英伟达权杖的这个敏感节点,认识他二十多年的华尔街顶级科技投资人 Gavin Baker 公开喊了一段话,听着像老友之间的养生叮嘱,细品全是顶级圈层的公开站台,
Gavin Baker 直接把话挑明了,说自己一直告诉 Jensen,去锻炼身体、去看日出、去睡足八个小时,因为想要拥有一个健全的头脑,必须先有一个健全的身体,他还特意补了一句,自己认识黄仁勋已经二十多年,一直把他当成真正的朋友,而自己是一个忠诚的人,
说实话,在如今这个墙倒众人推、恨不得踩上一脚的舆论场里,能公开亮出二十年交情和忠诚这两个字,本身就已经是极具分量的表态了,
更有意思的是他最后对黄仁勋年龄的判断,他直言自己完全不觉得 63 岁就到了终点,
在硅谷和华尔街眼里,这只是一个年轻的 63 岁,未来他还有大把时间可以为美国、为 AI 和整个科技界做出更多贡献,
以前我们总在新闻里看到那个每天只睡四五个小时、凌晨两点还在回邮件、穿着皮衣在全世界连轴转的超级工作狂,
大家习惯把他当成一台永不磨损的算力机器,却忘了再精密的系统在连续超频运转几十年后,也总有被迫强制冷却的一天,
但转念一想,这场突如其来的风波何尝不是一次命运强行按下的暂停键,离开日复一日的高压董事会和聚光灯,
真正去睡足八个小时、去看看清晨的日出,把透支了半辈子的身体重新养回来,
毕竟在科技史上,63 岁从来不是谢幕的年纪,只要身体和头脑还保持着巅峰状态,
今天交出去的权杖,未必不会在未来的某一天以另一种方式重新握回手里。
卧槽,这踏马太毒了
DeepMind 掌门人、诺奖得主 Demis Hassabis 刚刚把延续了上百年的现代学校教育底裤彻底扒光了:
全球几十万名数学老师每天关在办公室里,为同一门中学数学课重复手搓着一模一样的 PPT,这简直是人类智力资源最荒诞的工业级空转。
很多人至今还把名校名师的课堂讲授当成稀缺资源,拼了命去抢学区房和重点班,
但 Hassabis 直接用最冷酷的第一性原理砸碎了这个幻觉,
标准化的知识灌输从来不配占用昂贵的线下课堂,死记硬背和刷题完全应该被踢出教室,交给能按每个孩子毫秒级反应速度实时调频的私人 AI,加上全世界最顶尖的那几个大师录好底层大课,就足以把全球基础教育的讲授环节一次性彻底平替。
过去两百年里,学校本质上是一座把千差万别的碳基大脑强行塞进同一条流水线、按同一个节拍器拧螺丝的工业时代兵营,
老师沦为了人肉复读机,学生被困在枯燥的题海里比拼谁更像一台劣质硬盘,
而一旦 AI 和顶级大师把标准化的知识传递彻底接管,线下教室唯一剩下的合法存在价值,就只剩下了机器永远无法通过权重计算出来的五个维度:
人际博弈、复杂项目实战、创业野心、创造力,以及最致命的判断力。
这里有一个极度冷酷的反直觉真相:
当个性化 AI 把考高分的边际成本砸穿地板,按标准答案培养出来的做题家,在出校门的那一刻就已经直接沦为了算力废料。
未来的阶层分化,从来不再取决于谁的脑子里背下了更多现成的公式,
而是取决于当 AI 把一万种完美方案同时拍在桌上时,谁拥有拍板做决定的判断力,谁能攒起一个活人团队去把事情干成,
所以别再逼着孩子在死记硬背的泥潭里跟硅基芯片比拼记忆容量了,当流水线式的教书匠和做题家同时迎来历史性失业,谁能先把教室变成训练人性直觉与创业野心的实战沙盘,谁才能在算力时代拿到唯一不被平替的入场券
Itzul Girón retweeted
🚨 زلزال في الأوساط العلمية: جوجل تؤتمت شهادة الدكتوراه والبحث العلمي بالكامل بنسبة 100% بدون أي تدخل بشري! 🔬🧠💥
فريق البحث في Google يطلق رسمياً إطار عمل ذكي يُدعى "ScientistTwo".. نظام متعدد الوكلاء (Multi-Agent) ينفذ دورة البحث العلمي كاملة من الفكرة إلى الورقة البحثية المحكمة!
كيف يعمل هذا الوحش البحثي بشكل مستقل تماماً؟
💡 تعطيه المشكلة أو التحدي العلمي فقط.. وهو يتكفل بالباقي:
📚 يقرأ الأوراق العلمية السابقة ويحدد النقاط المرجعية (Baselines).
🧪 يصيغ فرضيات جديدة مبتكرة ويكتب الأكواد البرمجية ويجري التجربة بنفسه!
يقوم بدراسات الاستبعاد (Ablation studies) ويصمم نظام مراجعة أقران وهمي (Peer-review) ليجادل نفسه ويحسن ورثته البحثية قبل الصياغة النهائية!
النتائج الميدانية صدمت مجتمع العلماء:
🎯 تم اختباره مقابل أفضل أوراق علمية مقبولة في مؤتمرات نخبوية مثل NeurIPS و ICML.
⚡️ تفوق وحسّن نتائج أفضل الموديلات البشرية في 86 task مختلفاً!
📈 حقق قفزة في الأداء بنسبة بلغت 25.2% مقارنة بأعظم الإنجازات البشرية!
مؤلفات خالية 100% من الهلوسة المرجعية، مع أكواد تعمل بكفاءة وورقة علمية جاهزة للنشر الفوري!
المرحلة التي كنا نستخدم فيها الذكاء الاصطناعي كمساعد للتلخيص أو كتابة الكود انتهت رسمياً.. نحن أمام "باحث مستقل" يوسع حدود المعرفة البشرية!
إذا كان الـ AI يصيغ الفرضيات ويختبرها وينشر اكتشافاته بمفرده اليوم.. فكم يلزمنا من الوقت ليفتتح علماً جديداً لا نملك حتى الرياضيات لفهمه؟!
احفظ التغريدة فوراً، واقرأ الورقة البحثية كاملة عبر الرابط الرسمي بالأسفل! 🧵👇
Our latest model advances bring stronger capabilities for building and migrating software in low-resource programming languages such as COBOL, Fortran, and Visual Basic.
These languages power software that banks, governments, and businesses depend on. Progress here means understanding decades of business logic and helping developers modernize those systems while preserving what works.
We’re building these capabilities around the challenges developers and enterprise customers bring us. If your team is working on these systems, tell us which languages and workflows we should tackle next.
openai.com/index/low-resourc…
Itzul Girón retweeted
México vive una paradoja educativa frente a la IA:
83% de los universitarios ya utiliza #IA generativa para estudiar (Stanford HAI, 2026), pero alrededor de 76% de estudiantes y docentes de instituciones públicas no conoce reglas institucionales para usarla en sus universidades (ENIAG, 2025).
Es decir: altísima adopción, pero muy baja institucionalización.
De esto y más conversamos en el panel “Rediseñando la educación: el desafío de la IA”, junto a @itzulgiron, CEO de Celestial Dynamics, y @sissimpm, fundadora de The Dot Network.🤖📚
#ComisiónIA
Itzul Girón retweeted
Se acuerdan como hace meses aquí en México decían como jamás se iba a usar IA para programar por qué no servía para ambientes productivos
Just watched this "I'm done with AI" video by Brett Codes.
It seems some are having a hard time dealing with the realization that we are no longer "software" engineers but "product" engineers, and our job is fundamentally different.
It's a trend I've noticed in the past 2 years where entrepreneurial folks with technical backgrounds loved AI from the very first moment, whereas tech nerds who know 10 different programming languages in depth despise AI and hate using it.
When the entrepreneurial bros talk about what they're building they don't talk about tech stacks, languages or frameworks, they talk about features and how the features monetize and why users love them and how they'll lead to more users which monetize more and will keep the money printing machine running.
That's why to them AI is a bless because it removes the whole boring tech part which is only a means to an end.
But for us computer nerds it was the other way around, we'd look for a company that needs code written and we would get to write the code, that was our end, writing the code. Because we liked it, it was entertaining, we felt smart reading many times in a row that optimized function we wrote...
But really we never cared about the features or the users, we only made sure correctness is in place and the software isn't broken. We cared about writing the code, seeing it in the editor with our preferred theme and syntax highlighting, seeing the variables names we liked, the struct names we liked, the interfaces which created the perfect abstractions... it was all about the code.
If you take the code away from someone like that and leave only the product level work, i.e think about what features to implement, put yourself in the shoes of the user, QA test the result of the AI generated code and iterate on the _feature_ not the code... then it is pretty obvious they'll feel purposeless and depressed.
I was on that side, I've always looked at code as if it was something special, but somehow I've adapted to purely leading agents through features. I can't say I'm loving or enjoying this AI software engineering world, I had much more fun writing the code in the years before it.
But I don't hate it either, it doesn't make me depressed or purposeless, mainly because I know I'm the only one who can produce the right sequence of words in a prompt to get agents to do what we do at our company. You might think everyone can, but that is not true.
Remember that anecdote, a machine breaks and the mechanic comes to fix it, he looks at it and replaces one screw. Then he bills $10,000. When asked why replacing a single screw is so expensive, he breaks down the bill:
$1 for replacing the screw
$9,999 for knowing which screw to replace.
That is your position in software, with the only difference that you no longer replace the screw manually, you tell Claude to do it. But that doesn't change the fact that no one else knows which screw to replace.
Yes, someone might build a working product with no-code AI tools, just like anyone can fix a pipe by watching a tutorial, but there's only so much you can do before you actually have to call a plumber. At some point shit gets serious. And if you're actually a good software engineer, you're likely already working on serious shit.
The software engineering we knew is gone, and the new one is very different, but I suspect it will highly reward those who can adapt. People who could bridge the gap between the deep nerdy tech world and the non-technical user facing world were always very valuable, and I think AI will increase that even further.
The faster you can detach yourself from the code, the better things will go for you.
The "no AI coding" will essentially be a very small artisanal niche. How many companies do you know that use C to build optimized native desktop apps? Pretty much none because it is way faster and cheaper to use Electron and ship React to all platforms and devices.
That^ will happen with AI vs "no AI".
Itzul Girón retweeted
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 Kimi.com, Kimi Work, Kimi Code, and the Kimi API.
Open Weights by July 27, 2026.
🔗 API: platform.kimi.ai
🔗 Tech blog: kimi.com/blog/kimi-k3
Itzul Girón retweeted
And another open-weight release. Nemotron 3 Ultra has an ultra impressive capability:efficiency ratio!
Design-wise, it carries forward the Mamba-2-attention hybrid stack and LatentMoE introduced in the previous Super variant. But everything is a bit bigger.
Itzul Girón retweeted
Oigan sigan a mi amigo @EnriqueReyes_MX es una de las personas más talentosas que conozco para la administración de activos .
Alphabet $GOOG acaba de hacer algo que no hacía desde 2005: emitir nuevas acciones.
¿Por qué ahora? ¿Qué significa para los inversionistas de Google?
Explico 🧵
Itzul Girón retweeted
The world’s first @nvidia Vera Rubin NVL72 server rack is here.
We’re thrilled to deliver the first working, liquid-cooled @Dell PowerEdge XE9812 for @CoreWeave.
Built for the next era of AI infrastructure. 🚀🤝
Itzul Girón retweeted
What if you could take three completely different model families… and distill them into one tiny model? 🤯
📜 Paper: arxiv.org/pdf/2605.21699
MOPD (Multi-Teacher On-Policy Distillation) has become a standard procedure in post-training. We already distill multiple specialized variants of the same model into a single set of weights.
But what if we could go further - and distill models from entirely different families? Turns out, it is possible.
Today we’re releasing a paper on cross-tokenizer distillation - our first steps in this exciting direction. 📄
We distilled Qwen3-4B, Phi-4-Mini, and Llama-3B into Llama-3.2-1B.
MMLU jumped from 32.05 → 46.32 when using multiple teachers. 📈
The team is now working on Nemo-RL integration so the community can try this method in their own settings. Plus, we are scaling experiments up. 🚀
🤖 Made with AI
Itzul Girón retweeted
🚨 THE AI COST CRISIS HAS STARTED.
Microsoft reportedly told engineers to stop using Claude because AI bills were exploding, while Uber says its entire yearly AI budget was already destroyed by April.
Itzul Girón retweeted
"Hoy tenemos una oportunidad de poder marcar una estrategia de Inteligencia Artificial, de alinear lo que es la educación, la ciencia y la necesidad de ocupar la tecnología como una herramienta que potencie beneficios para nuestra sociedad", mencionó @itzulgiron
La UNAM instala el Consejo Coordinador de Inteligencia Artificial (CCOIA) para que la tecnología avance con perspectiva humanista, dijo el #RectorLomelí.
También afirmó que la Universidad, como institución pública, autónoma y plural, asume la responsabilidad de analizar los avances tecnológicos y guiar su uso en beneficio del interés general.
Este consejo deberá propiciar que la inteligencia artificial sea comprendida no como una fuerza externa e inevitable ante la cual hay que adaptarse, sino como un ámbito de decisión pública y académica, regido por una perspectiva centrada en el ser humano.
Más información > bit.ly/4uCpXrL
@CCOIAUNAM
@Proyectos_UNAM
Itzul Girón retweeted
🚨 ULTIMA HORA: Claude Mythos acaba de romper un sistema de defensa de Apple valorado en 2.000 millones de dólares.
No lo hizo de la manera obvia. Encontró un vector de ataque completamente distinto al que Apple tenía protegido.
Los detalles son brutales:
> 5 días de trabajo
> Costo total: 35.000 dólares en tiempo de API
> El mismo exploit en el mercado gris cuesta entre 5 y 10 millones de dólares
El equipo entregó un informe de 55 páginas directamente en las oficinas de Apple.
La parte que más impacta: la protección MIE de Apple funcionó exactamente como estaba diseñada. Mythos no la rompió. La rodeó, envenenando los datos que ingería el chip M5.
El equipo rojo de Anthropic lo confirmó esta semana: esto no es una cuestión de recursos computacionales. Es defensa nacional.
❗️🚨 BREAKING: Researchers used Mythos Preview to find the first public macOS kernel memory corruption exploit on Apple's M5 silicon, they give a glimpse into Mythos say it’s really powerful.
Apple spent five years and an estimated several billion dollars building Memory Integrity Enforcement (MIE), the hardware-assisted memory safety system built around ARM's MTE. It was the flagship security feature of the M5 and A19, designed specifically to kill the entire memory corruption bug class.
Researchers from Calif built a working exploit in five days.
According to Apple's own research, MIE disrupts every public exploit chain against modern iOS, including the recently leaked Coruna and Darksword kits. Calif walked into Apple Park this week and handed over the report in person.
Full 55-page technical report drops after Apple patches the vulnerability.
Itzul Girón retweeted
I’ve always believed the No.1 application of AI should be to improve human health.
That work started with AlphaFold, and now at @IsomorphicLabs with the mission to reimagine drug discovery and one day solve all disease!
We are turbocharging that goal with $2.1B in new funding.









































