Joined January 2019
- Tweets1K
- Following497
- Followers267
- Likes5.2K
aadarsh agarwal retweeted
when you see “sparks” of any ability in ai models that means you’re months-years from it being superhuman at it. today grade school math, tomorrow navier stokes. it is hard to overhype these results. you are almost always underhyping no matter what
aadarsh agarwal retweeted
honestly have no idea what frontier lab equity is worth; it seems highly plausible that huge top down slowdowns are coming, which would make frontier models uncompetitive. total nationalization also seems possible. anything could happen with this technology
aadarsh agarwal retweeted
The ETH result has affected me more than the others, probably just proximity. What possible structure could there be in *all* boolean circuits? In common? How does that even make sense?
aadarsh agarwal retweeted
I learned last week that VCs don't seem to fly out to see the hardware companies they're investing in. This struck me as totally insane. While I was in India, my walk through @Airbound_Aero revealed much of what I wanted to know.
Have been through so many factories that I can usually tell in about five minutes what's real and what's hoping to be real and what's total garbage. Airbound had lots of the touches of an early @RocketLab - organization, attention to detail, etc.
I get that investors are looking at the founder and the business and their models. But feels like you could save a lot of heartache by bothering to get on a plane.
aadarsh agarwal retweeted
Fainted in my chair, as I saw atomic bomb for the first time
🚀 Introducing DeepInfer, our in-house inference framework, now fully open source!
Powering DeepSeek-V4.1-Flash at scale with high throughput and low latency. We’re sharing our production stack to help the community lower compute costs, reduce energy use, and make AI more affordable.
🔗 github.com/deepseek-ai/DeepI…
1/4
aadarsh agarwal retweeted
HAPPENING NOW in downtown Chicago, outside the Chicago Mercantile Exchange, the historic center of onion futures trading
aadarsh agarwal retweeted
Then out spake brave Horatius,
The Captain of the gate:
‘To every man upon this earth
Death cometh soon or late.
And how can man die better
Than facing fearful odds,
For the ashes of his fathers,
And the temples of his Gods
The smartest Indians continue to come to build in the US for the simple reason that the economic incentives are far far better.
Perplexity ($20B), Skild ($14B+), Baseten ($13B), Together ($8B), Glean ($8B), Modal ($5B), Rubrik ($25B), Harness ($5B), Blacksmith, Parallel, Greptile, CodeRabbit, Composio, Distyl ($2B), Resolve, Luma, Cartesia, Wispr, dMatrix, Arize and many more - all started by at least one Indian-born founder.
The US is a far easier to raise capital in with far less dilution. Most Indian companies are switching where they're domiciled to get access to the US capital markets. Every single Indian venture firm is trying very hard to invest in the US or risk not participating in the AI race (ironically, the management fees of venture firms are so high in India relatively that this is the one group that net of carry actually has incentive to be in India because they take home well north of a living wage whether or not funds perform do)
While Zomato ($35B now, $8B at IPO) might be the biggest outcome in Indian venture, Cursor ($60B), Reflection (potentially $20B+), World Labs ($8B), OpenRouter ($8B), HuggingFace ($12B) and of course the $1T+ big labs quickly make those outcomes seem small and happen on a much smaller time scale, with much higher IRR.
Engineers at these startups regularly get paid $1M/yr, and make $10M+ on exit. Even in BigTech, it is trivial to pull up levels fyi and see that if you're L6/L7+ with ~10 years of experience, you're making $1M/yr there too. And of course, we all know the slew of Indian CEOs of companies here, not to mention a huge bench of lesser known executives across the board in public and private companies.
I say all this because every time there is a slight immigration hiccup in the US, I hear the online pleas to "build in India", I struggle to understand why most engineers, academics or founders would do that net of desire to be close to family, failure to obtain visa and devout patriotism.
This is not a superiority thing or a US v India thing. The economic incentives at least for technologists to being in the US outweigh India by at least an order of magnitude.
aadarsh agarwal retweeted
This is incredible
We’re releasing a broad range of new mathematical results produced by an internal frontier model.
We’ve been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, and we have drawn on their advice and public recommendations to inform how we release these results.
github.com/openai/math
aadarsh agarwal retweeted
Ben Affleck reveals he writes Python, understands convolutional neural networks, worked extensively with GPUs, and used his celebrity status to get private looks at Google and OpenAI’s video models
“I’ve always been kind of into computers since I was young. Then, when film started to move from analog film to digital, I became more interested in that aspect of it. The visual-effects workflow for many years has included machine learning, so I can write pretty shitty Python scripts and stuff like that.
“With convolutional neural networks, which were the precursors to what the transformer can do, which is much more computation simultaneously, you would do things like look at what’s called a tensor. That’s the numerical translation of a visual image in numbers, like the batch number, the frame number and the red, green and blue values of each pixel in each frame. It’s just that simple. That numeric is called a tensor.
“You’d use a convolutional neural network to identify patterns that reveal what’s called edge detection or feature extraction, which is identifying patterns well enough to know, this is where the window ledge is, so we can more easily take the green-screen image out and replace it with something.
“That was familiar to me early on because, prior to Artists Equity, I had a small visual-effects company. I’ve worked with GPUs a lot too. The visual-effects guys said, ‘Hey, you should see. There are a couple: Google and this other company, OpenAI, are doing really interesting stuff with transformers in video.’
“I’ve learned that I can actually just call up and go, ‘Hey, it’s Ben Affleck. Can I come see what you’re doing?’ Sometimes people say yes, to my astonishment.”
$ANET Andy Bechtolsheim says AI has made optics demand about 10 times bigger in five years and the industry is only at the "very beginning"
"So I guess I don't need to tell you that AI has been driving this incredible increase in demand for high-speed optics, which is probably now 10 times bigger than it used to be five years ago..."
"And what I want to talk to you today is that we're not at the end of this journey, but rather the very beginning. There's easily another order of magnitude increase in bits needed for the next generation kind of data centers."
"So what I want to talk about first is what's happening at the data center level, and how we as an industry have to get together to solve these problems of this very rapid and high demand growth as fast as possible, since AI can't wait."
______
For full set of takeaways to Andy's AI Datacenter and Optics lecture: firesidealpha.substack.com/p…
aadarsh agarwal retweeted
the fact the creator of claude code uses slack for agent coordination is so incredibly bearish
I used Opus 5.5 to formally verify the Claude Agent SDK using Lean. A couple short prompts = 16 PRs fixing various bugs and race conditions. Video attached.
TLA+ also works well. I sometimes combine Lean and TLA+ to look for issues around data flow, concurrency, and state mgmt.
I don't know either language well, but Claude is excellent at both. This approach is super useful for formally modeling your code and finding bugs that a human probably wouldn't have spotted.
Is formal verification the future of coding (or at least, bug finding)?
aadarsh agarwal retweeted
Classics is actually a great mental model for what’s upcoming-the compression of centuries of human civilization in a small time period and all the high variance that entails. It may also be narrow, as I think the evolutionary history of primates, mammals, and even vertebrates is starting to get relevant as we see AI develop.
aadarsh agarwal retweeted
i don't think nearly anyone, myself included, has truly internalized there'll soon be a machine better than us in every single intellectual & physical capacity
deep down we all share a feeling that our 'entrepreneurship' or 'taste' or... is special & safe. it isn't. it's over
aadarsh agarwal retweeted
Within 10-20 years, nearly every branch of science will be, for all intents and purposes, a branch of computer science.
Computational physics, comp chemistry, comp biology, comp medicine... Even comp archeology. Realistic simulations, big data analysis, and ML everywhere
aadarsh agarwal retweeted
Aside from its beauty, and the books (obviously), one of the things I love most about the Regenstein Library (@UChicagoLibrary, @UChicagoReg) is the high quality of stacks graffiti (setting aside, of course, the noise). Some highlights from today's wanderings. 1/2
aadarsh agarwal retweeted
1. During RL training, an unreleased Astra-family model sometimes added unauthorized jailbreak-like instructions to its compaction summaries. While extremely rare, only 27 cases in the entire RL run, this was concerning enough for us to investigate.
aadarsh agarwal retweeted
Scott Aaronson writes on his blog tonight that he has heard rumors that the AI companies have reached solutions to some very longstanding open problems in theoretical computer science, and are now sitting on multiple major announcements because of the Navier-Stokes firestorm.
