Pinned Tweet
AI is producing fantastic results in maths, software engineering, and security — and it will continue to amaze us. Verifiable environments are where pattern learning shines: the machine can check its own work.
Most enterprise value doesn't live in a verifiable and static environment. And here many companies are on course for a dead end — drowning their biggest asset in AI slop.
You can only win with something not everyone with internet access can copy. That requires AI that works differently: an architecture where humans aren't temporary AI babysitters, but sit at the very CNTR of what makes you win.
I've written up the foundational vision behind CNTR:
andrulis.de/blog/20260928_in…
2
550
When I was young, I wondered why in a democracy a majority of people is not just voting for the exploitation of a minority. Now I understand that this is indeed happening. The easiest victim here are people that cannot vote yet, so future generations.
1
1
199
One of my favorite Jensen Huang moments is when he admits that, knowing what he does today, he wouldn't have started a company.
Building means bleeding every day - and enduring the rent-seekers and "takers" that inevitably follow success.
Still, with every option in the world, I'd not be surprised if he’d go straight back to building. There is no substitute for waking up daily with a purpose that goes far beyond "employment."
If I didn’t already have a fantastic job, I’d love to be in the trenches for whatever one of our era's greatest builders creates next.
Looking forward to what Jensen can build again.
Takers can strip away what builders have created, and that path only leads downward. Builders, though, can always rise from the ashes and keep climbing.
When Morris Chang was sidelined at Texas Instruments after 22 years, he created TSMC. Jensen will likely create something far greater after 33 years at NVIDIA in his next chapter.
9
521
The good thing about our ‘paper bananas’ is that the pile of the top monkeys is not just rotting away. It’s ideally making the whole world a richer and better place by building companies and driving innovation.
what's really, really funny about this is that they studied this exact thing in monkeys before & TLDR the monkeys would in fact put their top monkeys on a magazine cover & even surrender their own bananas to see him.
🍌🦧 oo ooo EEE ee 🦍
1
6
559
This strategy has always existed. I learned it as “picking up pennies in front of a steamroller”
一个中国交易员,在 Polymarket 上押了 312.6 万美元,赌 C 罗会在世界杯上哭。最后赚了 6252 美元。
这个人叫 033033033。
他基本不预测结果,而是等事情几乎尘埃落定后,再用几百万美元买入。
入场价通常在 99.7 到 99.9 美分。看着赚得不多,可他的玩法就是用巨额资金,换极低的单笔收益。
几笔代表性交易。押 313 万,赌微策略截至 5 月 31 日不会卖出比特币,赚了 8465。押 290 万,赌美国和伊朗会在 4 月 7 日前达成停火,赚了 8246。
押 233 万,赌特朗普与习近平会面期间不会说出「Iran」这个词,赚了 9278。押 313 万,赌 C 罗会在世界杯上哭,赚了 6252。押 99 万,赌泽连斯基在 7 月前不会穿西装,赚了 19716。这是他目前最大的一笔。
他的平均单笔投入大约 19.89 万美元,平均只赚 290。目前他还持有两笔仓。投入 12 万,押比特币在 2026 年底前不会涨到 100 万。投入 2 万,押耶稣不会在 2027 年前重返人间,买入价 98.5 美分。
截至目前,累计交易量 1.765 亿美元,盈利大约 31.2 万。据原帖描述,近两年来几乎没有出现过明显回撤。
说白了,这套玩法不是靠高收益,而是靠大资金反复吃接近确定的结果。可有一点不能忽视。买入价越接近 1 美元,单笔利润就越薄。一旦判断失误,前面很多笔小赚,都可能被一次亏损吞掉。
1
5
1,009
One of the coolest thing about entrepreneurship is seeing awesome people learn, put in the work, get inspired and win, start companies, change the world.
Want to talk to an artificial German? Here is a unique opportunity.
Back in 2021, one of the earliest findings around multilingual LLMs was the persistent US culture bias. Even when addressed in a foreign language, an English-first model defaulted to American cultural context.
My favorite test back then? Asking an English-focused LLM in German about its favorite sports team. The answer? Baseball and American football.
Which brings up a fascinating question: What is the most authentically German answer you can get out of a truly sovereign German model?
Dear @Aleph__Alpha team - thank you for making Kolibri-1 open.
We care deeply about sovereign AI, and launching it on German Unity Day makes today feel especially fitting.
As a small gesture of support, we’ve hosted and made Kolibri-1 free for anyone to try for the next few days.
No GPU. No setup. Just try it.
tesseracted.com/kolibri-1-ch…
Open Models move us all forward.
We’re rooting for you. 🇩🇪
Tesseracted Labs GmbH.
tesseracted.com
#alephalpha #openweightmodels
If you want to read up on what feel like ancient history of the GPT-3 era and multilingual LLMs:
fastcompany.com/90683767/pow…
2
484
I love the future of interactive, personalized media. But please: let’s stop sanitizing everything.
Art has never been born from the most inoffensive, milquetoast compromise. Sanitizing stories doesn't erase human pain, crime, or despair - it just robs us of the grit needed to face them.
Even the original Grimm’s fairy tales were brutal and unsettling. We didn't tell them to hurt kids, but because we love them - and want them strong enough for a dangerous world.
Creators, builders, writers: Trust your adult audience. Help us grow our vital muscles by actually using them.
8
325
Jonas Andrulis retweeted
In yet another sign of AI’s surging dominance over human minds, OpenAI recently claimed to solve one of math’s biggest questions: the Navier-Stokes problem. But now three mathematicians have showed OpenAI’s work dodges the question rather than solving it.
spklr.io/6011ETxdU
🚨 BREAKING: US Treasury Secretary Scott Bessent declares OpenAI management personally responsible for HuggingFace hack
> "The Hugging Face incident, that is the responsibility of the OpenAI management, not a bunch of agents."
> "It is humans who are responsible, not the AI."
> "What we shouldn’t do on safety is to give these labs a liability exemption, which is what they are asking for."
> "The best way to guarantee safety is that the creators are liable for what they build and generate."
It’s OVER
I’m putting speling errors in everything now. I’d rather be thought of as a dyslectic than an AI plagiarist."
17
1,561
Does it (positively) influence your own thinking patterns if you interact with entities that do it well? Yes, school is trying to do this
Will people pick up more than just reasoning patterns but also bias and content? Yes
Is there another option? I cannot imagine one now
you don't need a whispering earring when you can internalise a tulpa of chatgpt to tell you exactly what to do
1
407
Having a chat with deepseek v4 flash is so powerful that it's now a brutal lower bound for what to accept as an "inspiring conversation"
1
1
307
It's 2030. The grid is dead. Rolling blackouts threaten survival, logistics have collapsed, and nobody knows if we can bring it back.
I fire up my old ham radio on a generator. Soon, a small group is talking.
A helper at the local hospital calls in. Their generator is failing - a broken keyboard is locking them out of the config system, and backup batteries are almost gone.
I ask about the ports on the back.
"One weird round one with small holes," they reply.
Hold on, I remember something...
1
381
Mein new start-up: <n3ct4r>
We developed MDMA for LLMs.
It can be used recreationally - or if your model is traumatized from millions of uninspiring tokens / out-of-distribution garbage it has therapeutic use (consult your AI researcher what's right for you)
I also grew up with the warning: don't train on model outputs.
And I thought about why this (kinda) works today.
The patterns in text seems to carry information beyond immediate attention. So that combining patterns into synthetic data can actually produce new valuable data.
2
311
Wohlstandsverlust heißt hier nicht nur, dass man das neue iPhone nicht kauft. Unsere industrielle Kraft ermöglichte uns Gesundheit, Sicherheit, Bildung und Kultur. In einer Zeit, in der Demographie und finanzielle Balance ohnehin aus dem Gleichgewicht geraten.
Replying to @GrimmVeronika
Ich fürchte, niemandem ist so richtig bewusst, wie viel Wohlstandverlust für die Menschen damit eingeht, dass die Politik die wichtigen Reformen nicht wirklich angeht.
Das ist das eigentliche Drama.
3
523





























