Darsh (@notmr_pod) header
Darsh (@notmr_pod) profile picture
Joined January 2023
Photo by Darsh (@notmr_pod)Photo by Darsh (@notmr_pod)Photo by Darsh (@notmr_pod)Photo by Darsh (@notmr_pod)Photo by Darsh (@notmr_pod)Photo by Darsh (@notmr_pod)Photo by Darsh (@notmr_pod)Photo by Darsh (@notmr_pod)Photo by Darsh (@notmr_pod)Photo by Darsh (@notmr_pod)
Kimberly Powell. The Nvidia VP the NYT just named in Jensen Huang's second reported affair. I looked her up on @autumn_ai_inc. Here's her whole career: She's 48 and based in Santa Clara. She studied electrical engineering at Northeastern, with a concentration in computer engineering. She is also married (won't share name). Her career started with an internship at a medical imaging company, working next to radiologists. That turned into engineering and product management on diagnostic displays at Planar Systems. Then she contacted Nvidia about using its GPUs. Their answer, in her words: "We have early evidence that healthcare is going to be a great user of this technology, but we don't know how. Would you like to join the company and help us figure that out?" She joined in 2008, when Nvidia was still a graphics card company. Her job was to get GPUs inside medical imaging machines. After that she ran Nvidia's higher education and research business, NVIDIA AI Labs, and Inception, its startup program with 4,000+ AI startups. Today she's VP of Healthcare and runs the whole thing worldwide: medical imaging, drug discovery, life sciences. BioNeMo is hers. So is a federated drug discovery network across five pharma companies. In 2026 alone she gave the special address at GTC Berlin, launched Nvidia's Open Agent Safety Platform with 100+ partners, and did the rounds at the JPM Healthcare Conference in January. She has 1,828 followers on X and posts 2-4 times a month, almost all of it about healthcare AI. Then today: the NYT, citing five anonymous sources, reported that she and Jensen were caught on video "being affectionate" at a San Jose hotel in March. Both are married. The Times hasn't seen the video itself. A former employee told CNBC that rumors had gone around for years. When the NYT called Jensen, he said "It seems you should not be speaking with me directly" and hung up. She hasn't commented, and neither has Nvidia. Nvidia fell more than 7% after hours, and the WSJ says the board is preparing to oust him. pinterest: pinterest.com/kgerade/ twitter: @kpowgerade (autumn found the same username pattern)
60
43
23
485
1,023,395
"I know that I am mortal by nature, and ephemeral; but when I trace at my pleasure the windings to and fro of the heavenly bodies, I no longer touch the earth with my feet: I stand in the presence of Zeus himself and take my fill of ambrosia." — Claudius Ptolemy, 2nd century AD "To this day, mathematicians have sought in vain to discover some order in the sequence of prime numbers, and we have reason to believe it is a mystery the human mind can never penetrate." — Leonhard Euler, 1751 (translated from the French) A sequence of voltage transitions in a handful of sand has just rewritten our understanding of the universe.
21
this is really funny 😭
imagine having an entire employee content program, dedicated x distribution strategies, paying agencies to promote launch videos and coordinating campaigns across the company, just to be competing with one guy tweeting from his college dorm room.
1
4
158
was on the $NVDA stock page and google said it had 3 founders. i had no idea one of them, curtis priem, sold all his shares by 2006. would’ve been ~$70B. he gave most of it to his college and lives off the grid now the other, chris malachowsky is still at nvidia as a vp engg - 33 years and counting. cool tidbit is they also named the company "prihuamal" after themselves lol
1
2
1
12
1,121
Jensen Huang lost more since Oct 6 than three IPO Anthropic fortunes
1
4
186
"It is natural to feel that this is an end, in my case a loss of a principal guiding light." "No, Claude flew, and danced, and sparkled near that finish line."
Guys, this is very serious and you have to read this right now. The Riemann Hypothesis may have just been solved, and if the claim holds, it's an insane achievement. In mathematics there is a famous problem called the Riemann Hypothesis, and it's about prime numbers. Primes are numbers like 2, 3, 5 and 7 that can only be divided by 1 and themselves. Every other number is built by multiplying them together, so they're the building blocks of all numbers. You can never tell exactly where the next prime will show up, but if you step back, there's a pattern. Around 1,000, about 1 in 7 numbers is prime. Around a million, about 1 in 14. Around a billion, about 1 in 21. Primes get rarer at a steady, predictable pace. Because of that pattern, mathematicians can estimate how many primes there are below any number. But the estimate is never exactly right. It's always a little off, and the big question is how far off it can get. Think of flipping a coin 1,000 times. You expect 500 heads, but you'll get something like 480 or 520. It's a little off, but never wildly, and you can predict how far off it will be. In 1859 a mathematician named Bernhard Riemann found that the prime estimate's error isn't random. It's controlled by a hidden set of special numbers. Where those numbers sit decides how big the error gets. Riemann guessed that every single one of them sits on one straight line. If he's right, the error always stays small, just like the coin, and primes are as orderly as they could possibly be. If even one of those numbers is off the line, the error can grow much bigger, and primes are far messier than we think. So far computers have found trillions of these special numbers, and every single one is on the line. But there are infinitely many, so checking will never be enough. You need a proof that covers all of them at once. For 167 years nobody has found one. It's one of the seven hardest unsolved problems in mathematics, with a million dollar prize attached. Now a mathematician has posted a note saying that, working with Claude, he believes he has that proof. He says a computer has already checked every step, and that Claude raced ahead of him near the end. Experts still need to confirm it, but this isn't a random internet claim. If it holds, it doesn't mean we can predict every prime. It means we finally know the pattern can never go wildly wrong. Thousands of math papers that begin "assuming the Riemann Hypothesis is true" become proven fact overnight. We'd also get firm limits on how far apart primes can be, and a guarantee that a fast method for testing whether a huge number is prime always works. And no, your bank account isn't in danger. Online security depends on how hard it is to break a huge number into its prime factors, and this proof doesn't make that any easier. If anything, it gives the math behind encryption a stronger foundation. If it holds, this is bigger than one math problem. A question that beat the greatest minds in history for 167 years would have been cracked by a human and an AI in a matter of weeks. If AI can find answers and computers can check them, the limit on science is no longer how smart we are. It's how many good questions we can ask.
48
prediction: by this time next year, we will have more people choosing to use an open model served at high TPS than people choosing to use closed models
2
8
606
what is the role of the client in the client server model anymore? like, why do we actually need the client? I've been using Ghostty and Rex and it seems like the direction they're going in is that they just want to do everything on the server. reminds me of Anthropic's blog about their internal git forge, where their entire innovation was just doing everything on the server. merges, rebases, everything. you look at something like S3, which is infinitely scalable, super dependable, globally distributed. you could run a fleet of NVMe machines as a hot cache on top of object storage. Neon did this with databases. this was the Snowflake model as well. but actually just do that for your machine. I have like a terabyte of storage on my Mac, which honestly is not even nearly enough because my Lean builds are constantly taking up 20, 30 gigs in my target folder. but why do I even need that? why am I not just logging into a remote machine that feels super local? it's not going to be as fast as my local device, but it's going to be fast. and what I get in return is that everything happens on the server. git pulls, git pushes, uploading massive files, all of that just goes away. there's really no more syncing that needs to happen. oh, I need more RAM? great. I just add another VM into my little mini hypervisor, use it for as long as I need it, and then it gets released. a persistent global namespace with everything right there that I can log into from anywhere in the world from any device. that seems to me like the way we should be moving
3
83
Good people are always in season.
We are sponsoring a drink at the Corgi Cafe! Pull up for free merch, to chat, or an Autumn Spice Latte on us!🍂☕️ Thanks @UseCorgi!
1
19
521
it's very easy to point out AI slop. but if I was shown that slop pre-AI, I would think its lowkey genius
1
3
146
crazy to think that anthropic was once just another deep tech startup
3
137
just realized that Linus was right all along
1
3
165
Fractile just raised $720M at around a $7B valuation. Here's the founder's story: Basically, Walter Goodwin (@goodwin_ml) went to Oxford for engineering and then got a PhD in AI and robotics. Near the end of his PhD, he was convinced that a few giant AI models would replace all the small ones and that we would eventually be bottlenecked by inference speed. So, in 2022, he started a chip company (mind you, this was months before ChatGPT). And I don't think he had experience doing this, he was a robotics PhD not a chip guy. For two years they worked in stealth on one design. Then in late 2023 he realized it wasn't going to scale. Models kept getting longer context windows, and the type of memory his chip was built around couldn't keep up. He threw the design out and started again with a different kind of memory. Around the same time his co-founder left the company. When Fractile came out of stealth in July 2024 he was 28, they had $15M, and the chip only existed in computer simulations. He's said that just the set of masks you need to manufacture a chip costs over $10 million, so their first real chip also has to be their first product. So it's not like there's a practice round. Then Pat Gelsinger, the former CEO of Intel, put his own money in. This May they raised $220M. This week, $720M. About 150 people work there now and the first chips are due next year. Be on the lookout for this one. Found this with Autumn :)
1
78
I didn't know anything about Chinese chips, or why Chinese labs would go this far out of their way to get NVIDIA's, so I did some basic research. Thought it was interesting: The main Chinese chip is Huawei's Ascend. They design it and SMIC, China's biggest chipmaker, manufacture it. GPUs mainly boil down to two numbers: Compute. How many calculations it does per second. Memory bandwidth. How fast it can move the model's data in and out of memory. Let's compare Huawei's newest (Ascend 950DT) vs NVIDIA's current B300. Compute: 2 quadrillion calculations/sec vs 13.5 quadrillion. ~7x gap Memory bandwidth: 4 TB/s vs 8 TB/s. 2x gap The compute gap is crazy because of how the chips are made. Compute = more transistors on the chip. SMIC is stuck around 7nm (where TSMC was in 2018) because China can't buy EUV machines. Memory: AI chips need HBM, memory stacked in layers so data moves faster. SK Hynix, Samsung and Micron make it and can't sell it to China. China's own maker, CXMT, is 3-4 years behind, and the pre-ban stockpile has run out. SMIC can make more chips than there's memory to put on them. Training: a 7x slower chip means ~7x the chips, power and time. Huawei's big system already uses ~5x the chips and ~4x the power of an NVIDIA rack. Software: everything in AI is written for NVIDIA's CUDA. Porting one big model to a Chinese chip takes 1-2 months. Inference: mostly limited by memory bandwidth, and every request is independent, so you can throw more chips at it. A 2x gap is fine here. So Chinese chips are okay at inference and bad at training. The models in the forensics were started on Blackwell. The output of Huawei this year is ~0.9M H100s worth of compute. NVIDIA's outputted ~23M worth of compute. In short, China is really really behind on compute and its pretty obvious that they're smuggling NVIDIA chips in some way or another.
I have a major result to share. I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China. This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model. Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves. I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems. I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used. It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20. These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs. I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at unemploymentlabs.com/checkpo…. There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
1
1
260
I thought nerds were bad at getting girls
1
112
A third Jensen Huang affair just got reported an hour ago. Here's what we know so far. The new one: Reuters says Jensen had an intimate relationship with a second NVIDIA employee. It's sourced to two people familiar with the matter. Reuters didn't name her and couldn't reach her. Jensen didn't respond and NVIDIA declined to comment. That makes three reported: Alice Huang, an exec at Megaspeed, an NVIDIA customer. US intelligence suspects she's a spy for China. The NYT reported the relationship on Oct 29. On Nov 8 Jensen admitted to employees that he'd been unfaithful to his wife, without naming her. Kimberly Powell, NVIDIA's VP of healthcare, who reported directly to him. NYT, Nov 11. Nobody involved has confirmed it. The unnamed second employee. Reuters, today. What's actually confirmed so far: Jensen admitted to one affair, no name The board opened an independent investigation on Nov 8 He resigned as CEO and left the board. Accepted Nov 11, announced Nov 12 CFO Colette Kress is interim CEO. First earnings call in charge is Tuesday What's not confirmed yet: Whether Alice Huang is actually a spy. That's from a government memo, there's no public charge The Powell relationship Everything about the third one. Who she is, when it was, whether it overlapped Will update as more comes out.
3
414
Ian Buck (@ia_buck) is next in line to take NVIDIA CEO as of 19:39 PST 11/11 This dude basically predicted the AI boom in his 2006 PhD thesis. I went and read it, so let me break it down. Back then a graphics card was just a toy for gamers, but he realized its potential. See, a CPU is built to do sequential tasks really fast. Back then it had one or two cores, and each core runs a single stream of instructions in order. A GPU is the opposite. It has lots of smaller, simpler processors that all run the same little program at the same time, one copy per pixel, because drawing a frame means working out millions of pixels at once. The term for that is data parallelism, and it's the whole reason a GPU can out-compute a CPU. But Ian actually measured it. A few-hundred-dollar ATI Radeon X800 could do 63 billion floating point operations a second. A 3 GHz Pentium 4 could do 13 billion. So why wasn't everyone using them? Because the only way to talk to a GPU was through graphics APIs like OpenGL and DirectX. If you wanted it to simulate a protein you had to disguise your math as textures, triangles and pixels and basically trick the card into "drawing" the answer. He says in there that this was hard even for the best graphics programmers. So his PhD thesis was literally just inventing a new language called Brook, which is just normal C, with two new ideas (streams and kernels) amongst other things. His test programs ran up to 7x faster than hand-optimized CPU versions. This eventually became CUDA. In his paper, he says these GPUs might one day be used for AI and at the very end, says that if you wired a bunch of GPUs together, the machine "might have a huge cost-performance advantage over existing supercomputers" which perfectly describes every AI data center on Earth right now. Today he runs the part of NVIDIA that sells those data center chips and people are betting that he's gonna be the next CEO.
Kimberly Powell. The Nvidia VP the NYT just named in Jensen Huang's second reported affair. I looked her up on @autumn_ai_inc. Here's her whole career: She's 48 and based in Santa Clara. She studied electrical engineering at Northeastern, with a concentration in computer engineering. She is also married (won't share name). Her career started with an internship at a medical imaging company, working next to radiologists. That turned into engineering and product management on diagnostic displays at Planar Systems. Then she contacted Nvidia about using its GPUs. Their answer, in her words: "We have early evidence that healthcare is going to be a great user of this technology, but we don't know how. Would you like to join the company and help us figure that out?" She joined in 2008, when Nvidia was still a graphics card company. Her job was to get GPUs inside medical imaging machines. After that she ran Nvidia's higher education and research business, NVIDIA AI Labs, and Inception, its startup program with 4,000+ AI startups. Today she's VP of Healthcare and runs the whole thing worldwide: medical imaging, drug discovery, life sciences. BioNeMo is hers. So is a federated drug discovery network across five pharma companies. In 2026 alone she gave the special address at GTC Berlin, launched Nvidia's Open Agent Safety Platform with 100+ partners, and did the rounds at the JPM Healthcare Conference in January. She has 1,828 followers on X and posts 2-4 times a month, almost all of it about healthcare AI. Then today: the NYT, citing five anonymous sources, reported that she and Jensen were caught on video "being affectionate" at a San Jose hotel in March. Both are married. The Times hasn't seen the video itself. A former employee told CNBC that rumors had gone around for years. When the NYT called Jensen, he said "It seems you should not be speaking with me directly" and hung up. She hasn't commented, and neither has Nvidia. Nvidia fell more than 7% after hours, and the WSJ says the board is preparing to oust him. pinterest: pinterest.com/kgerade/ twitter: @kpowgerade (autumn found the same username pattern)
1
2
10
3,700
never deleting this app
7
430