I’ve worked that heatshield my entire adult life. Peak career moment today
To go from little 7 tile patches to a full reusable ship. Absolute marathon
I’m so grateful I got to do something useful, that I was given the chance
I don’t really have many good memories from this experience.
It was so difficult. It almost looks easy now, but none of it came easily
Please do something amazing with this rocket. This tech is such a gift, a love letter to the species
Philippe Back retweeted
Classic three bars down off OP= sign of trend day potential. Initial downside = Pocket area.
Philippe Back retweeted
The revenge arc is going to be insane
Jensen Huang, CEO Terafab
Nice to see @elonmusk move here.. welcome to SpaceXSI Jensen.
Mark this as the moment when NDVA losses become SpaceX wins
Philippe Back retweeted
미국 주식 하루 거래량의 25%를 혼자 돌리는 '월가 최종보스' 시타델의 비밀 🤯
시타델(Citadel) 창업자 겸 CEO 켄 그리핀이 시장조성자(마켓메이커)인
시타델 증권이 미장에서 5,000만 달러(약 700억 원)짜리 매도 폭탄을 어떻게 받아내는지 직접 썰을 풀었음.
• 5,000만 달러 매도 폭탄 방어 원리
큰손이 5,000만 달러어치 주식을 한 번에 시장가로 던졌는데,
정확히 그 찰나의 순간에 받아줄 매수자가 시장에 없으면 원래 주가는 호가창 뚫리면서 비정상적으로 떡락해야 정상임.
근데 마켓메이커가 중간에 개입해 과도한 가격 왜곡을 막아줌.
• 선 체결 후 매수자 찾기 (자기 자본 투입)
시타델 증권이 자기네 자체 자본을 먼저 태워서 매도 물량을 즉시 체결시켜 줌.
한마디로 일단 자기 돈으로 물량을 다 받아주고,
나중에 시장에서 그 주식을 사 갈 최종 매수자가 나타날 때까지 기다리며 넘기는 구조임. (대신 호가 스프레드 챙기고 가격 변동 리스크도 본인들이 떠안음 ㅋㅋ)
• 미국 증시 거래량의 약 25% 점유
켄 그리핀 피셜,
현재 미국 주식 시장에서 매일 거래되는 전체 주식의 거의 25%를 시타델 증권이 처리하고 있음 ㅎㄷㄷ.
미국 증시에서 거래되는 주식 4주 중 1주는 시타델 손을 거친다는 소리임.
우리가 미장 할 때 주문 넣자마자 슬리피지 크게 안 나고 0.01초 만에 체결되는 게 다 얘네가 유동성 윤활유 역할을 해줘서 가능한 거임. 동시에 미국 증시 거래량의 4분의 1을 단일 기업이 쥐고 있다는 게 얼마나 어마어마한 영향력인지 보여주는 대목이기도 함.
Not financial advice
Philippe Back retweeted
Gemini 4 Carbon is here!
Our strongest model is now available, with frontier performance in cybersecurity and long-horizon capabilities. Built for your toughest engineering work, Carbon takes the time to explore thoroughly, reason deeply, and run extended verification loops to tackle your most complex workloads with less prompting and hand-holding.
Alongside Carbon, we’re previewing /swarm in Google @Antigravity so you can spin up a team of agents from one prompt, assigning Carbon to architecture and auditing while parallel agents build and test every change.
Here’s how you can build with Carbon ↓
Philippe Back retweeted
The guy in the most famous engineering video ever made had no idea what he was saying. None of those words are real, and they've been fooling people since 1944.
His name is Mike Kraft. He had a tape recorder hidden in his pocket with a wire running up to his ear, and he just repeated whatever it told him with a totally straight face. Honestly, it's the most accurate vibe coding video ever made, and he shot it decades before anybody had heard the phrase.
The script goes back way further than the video.
In December 1944, a British engineering student named John Hellins Quick wrote a fake article for the Institution of Electrical Engineers' student journal. He gave his made-up machine hydrocoptic marzlevanes and an ambifacient lunar waneshaft, and assured everyone that side fumbling had been effectively prevented. That was three years before the transistor was even invented.
Time magazine ran it in 1946, and some readers bought it. One wrote in asking what a dingle arm was, and the editors answered him with more made-up jargon. Another guy in Hoboken figured he'd be able to pick one up at a Manhattan department store any day now.
Then in 1962, engineers at GE snuck a data sheet for it into the official General Electric Handbook, laid out exactly like the real products on the pages around it.
Around 1977, a Detroit actor who made industrial training films, Bud Haggart, talked a GM crew into staying late after a real shoot so he could put it on camera. Kraft had worked with Haggart, and later did his own version for Rockwell Automation, standing in front of an Allen-Bradley motor control center. It went out on the company's end-of-year VHS tape and he figured that was the last anyone would see of it. Then somebody put it on eBaum's World, and it's been bouncing around the internet ever since.
He's made basically nothing off it. He told Ampere News his fame is so low that "bugs can crawl under" it.
His theory for why it never gets old is that people see themselves in it. Anybody who's really deep in a field sounds exactly like this to the person sitting across from them, and the video lets you hear it the way they do.
Kraft didn't actually memorize the monologue until more than 30 years after he filmed it. He only did it because he was scared somebody at a conference would ask him to perform it without the earpiece.
That's more effort than most of us will ever put into understanding the code we shipped last week.
Omarchy is forever free. There's no monetization scheme coming. I'm already rich and the Omacom Foundation is a well-founded nonprofit.
The only ulterior motive here is to make Linux win on the desktop and create the perfect, malleable computer for all to enjoy!
Andrej Karpathy:
“A 1B model can be much more effective than most current LLMs.
90% of current models waste a ton of capacity and money on remembering stuff that doesn’t matter.”
here is his 5 core thesis about Small LLM’s:
1. the intelligence was never the knowledge. Strip the memorized internet, keep the reasoning. What’s left is the cognitive core, and it’s small.
2. the core might fit in ~1B params. Maybe less. “Plenty of room at the bottom.” We’ve been paying for a library when we needed a brain.
3. size stopped being the bottleneck. What makes an agent work is the tools it can call and how tight the feedback loop is, not parameter count.
4. intelligence lives in the system, not the weights. A raw LLM is a CPU with no OS. Small core + good harness > giant model alone.
5. you can own it. A model on your hardware is one no one can rate-limit, price-hike, or deprecate on you.
in 12 months the gap between teams shipping on-device and teams paying per-token frontier bills won’t be capability.
TwIL an 8.8B model at 90%+ of frontier isn’t a benchmark flex.
It’s the thesis Karpathy’s been repeating all year, made real.
study more on this below ↓
Modest intelligence is arriving faster than most people realize.
Today, we're open-sourcing TwIL-8B.
Just 8.8 billion parameters, yet approaching frontier-model scores on several difficult benchmarks:
• 93% of frontier performance on coding
• 92% on instruction following
• 90% on contest mathematics
All from a model small enough to run on your own hardware.
But the benchmarks aren't the most important part of this announcement.
For years, the AI industry has focused on building the most capable model, regardless of the cost required to run it.
I believe that equation is changing.
The next era of AI will be defined by the cost of completing work. Not simply the intelligence of the model, but how efficiently that intelligence can accomplish the task.
If a model with 8 billion parameters can reliably complete the same work as one with hundreds of billions, why pay for the larger model?
As smaller models become more capable, intelligence becomes something we can distribute, specialize, and run across the hardware we already own.
That's the future we're building at @thewebAI
TwIL-8B is available today. Open weights.
And this is just the first of several announcements we're making today.
HF Link: huggingface.co/webAI-Officia…
Philippe Back retweeted
JUST IN: Leopold Aschenbrenners Fund (Situational Awareness) has reportedly officially blown up…
These are the current (now liquidated) holdings:
- $MU bought at $1,153, now $703, (-39%)
- $CORZ bought at $25.6, now $11.1, (-56%)
- $SNDK bought at $2,268, now $874, (-61%)
- $NBIS bought at $273, now $121, (-55%)
- $CRWV bought at $99, now $42, (-57%)
- $BE bought at $301, now $165, (-45%)
Use him as a lesson why you NEVER over-leverage into high risk positions.
That concludes an end of an era…
Readers added context they thought people might want to know
Situational Awareness's public equity book was sold to Citadel in July 2026, not Nov 5. The firm hasn't closed: it retains a ~$5B Anthropic stake, continues operating, and 2026 returns are still reported around 80%. aiweekly.co/alerts/aschenb… withintelligence.com/insights/arche…
Philippe Back retweeted
Smalltalk-80 by the Book: fully working implementation in Smalltalk (in @pharoproject) of the Smalltalk-80 VM as published by Xerox PARC in 1983 github.com/ljulliar/smalltal…
Philippe Back retweeted
I've been coding for 40 years, and loved it. I don't code anymore, but I still love building things for the same reasons, and it provides the same rewards.
Coding just happened to be a means to an end for me, one I was good at, was lucky to get paid to do, and enjoyed the process of. Yet I don't miss it.
I spent about 3 years writing a LED management system that could deliver arbitrary graphics over WiFi to various LED devices, like this panel. But I had to drive it from a Pi.
They make a receiver card that works over Ethernet, so "all" I had to do was port my code to it!
I connected a Mac directly to a Colorlight 5A-75B receiver card to drive eight LED panels. The goal was simple: display solid blue.
Naturally, this became an investigation involving raw Ethernet packets, reverse-engineered vendor libraries, volatile configuration tables, slow-motion 240fps video, and a brief period during which I apparently owned only five of my eight panels.
The card had never been commissioned (or configured) for these panels. We had no Windows machine, and the receiver’s idea of sensible defaults was “green bars, with enthusiasm.”
We extracted the vendor’s configuration machinery, generated an FM6124 profile, and applied it to RAM without writing the card’s flash. Eventually, we achieved two bright blue panels and several dark ones.
Then I counted the panels again.
Eight. There were eight. I honestly thought I only had 5.
Correcting the geometry got us uniform blue. It also flickered like a nightclub operating under a questionable electrical permit.
The crucial clue: stop transmitting, and the picture became steady. Our packets were disturbing the peace.
Precise macOS scheduling cured the flicker. Then live content revealed a cyan seam and repeated sections. Sending smaller packets made the picture repeat more neatly, which was technically progress if you were building wallpaper.
The culprit was duplicated pixel packets. Sending each once fixed the image. Sending one synchronization packet instead of two removed the big backward jumps.
But occasional stutters remained.
We verified that 30 new images arrived every second, with about 50 frames buffered. Every image went out twice at 60 Hz, in order. The packet captures looked immaculate. The display occasionally looked as though it had forgotten why it entered the room.
So we built a test pattern with frame counters, binary codes, and moving shapes, then filmed it at 240 fps.
There it was: one image held for about 50 milliseconds, the next for only 16.7. A tiny pause, followed by a hurried attempt to look busy.
The final timing change was to upload the pixels early and reserve the precise refresh deadline for the sync packet alone. Sync intervals tightened from roughly 16.27–17.05 ms to 16.64–16.70 ms. Both the test pattern and live ticker visibly improved.
Solid blue turned out to require counting to eight, teaching a Mac punctuality, and explaining to an Ethernet receiver that saying everything twice does not make it clearer.
All the fun, none of the code :-)
Philippe Back retweeted
Ok this is hilarious!
I got this emailed to me anonymously (I think someone’s bot sent it) and I was on the floor laughing.
Philippe Back retweeted
Feels like the algo market making machines have been turned off as of yesterday - might be like this till after the elections. noise spikes even more thin and whippy = DO NOT CHASE.
Philippe Back retweeted
the 1st video of naijm, the company i am building with @daedalium, is out!
its very much introductionary but it'll help you understand what we do and how we think!
i am super proud and happy, the adventure is only starting!
youtube.com/watch?v=fK3Uukdt…
Philippe Back retweeted
Sony built an AI that discovers scientific facts on its own.
And it reportedly discovered actual previously unknown facts about "aging"
It's called "Hakken"
Hakken analyzes millions of research papers, maps how scientific facts evolve over time, and predicts exact, undocumented relationships that humans haven't found yet.
It generates a mathematical confidence score for scientific facts that do not currently exist in any literature.
Researchers pointed it at the biology of aging. The AI generated 1.5 million brand-new hypotheses.
But they didn't stop at the computer screen.
They took the AI's best predictions and physically tested them in a real-world wet-lab.
And the results are historic.
Physical lab confirmed two previously undocumented biological interactions in the human body (involving TP53, BAMBI, RAF1, and TNF).
the AI successfully saw the invisible mechanics of biology that human scientists had missed. And the lab proved it was right.
We are crossing a massive threshold.
We used to build AI to read scientific papers faster. Now, AI is writing the next chapter of human knowledge before we even run the experiments.
The bottleneck for scientific progress is no longer human intelligence.
It is how fast the physical labs can keep up with the machine.
Philippe Back retweeted
It's not just use of DeepSeek, it's also things made with DeepSeek and things that use DeepSeek.
If DeepSeek is a part of your SDLC, or if DeepSeek is a part of your SBOM, as an example - you can't do business with the US Govt
Wow, this is big, though not unexpected. US is removing DeepSeek from usage from all govt systems.
The 2nd/3rd order effects from this may be catastrophic though since all 3rd party systems need to be scrubbed of it and this may set a precedent/fear amongst almost every company for compliance.
And not just US companies, but any one that deals with any US company.

































