vinjai (@vinj_ai) header
vinjai (@vinj_ai) profile picture

building the @appliedcompute model factory / previously AI @Watershed, math @Stanford. opinions my own

Joined August 2025
Photo by vinjai (@vinj_ai)Photo by vinjai (@vinj_ai)
Like most mathematicians, I believed the Riemann Hypothesis to be completely out of reach at present (even with the best AI tools), because I expected it to require substantial theory-building, probably in arithmetic geometry. The kind of new theory-building that fills entire books, and that AI isn't so good at yet. Mathematicians have a notion of how deep a problem is. A sufficiently deep problem is generally intractable until enough of the theory is in place. If difficult problems are like valuable mineral reserves buried miles underground, you first need to build the infrastructure to reach them. What Levent's proof seems to have done instead is something completely different. It takes an incredibly complicated tour through the infrastructure built for a hundred other mineral reserves, situated far from the destination, and somehow create a wormhole to the original destination. This is not theory-building in the usual sense. It is something much weirder and more alien. The best human mathematicians know how to build theory, painstakingly, to reach a goal. But no human brain, on its own, is capable of creating a wormhole that requires a 25,000 page long tour through existing infrastructure with no obvious or realistic connection to the goal. But huge credit to Levent here: I think no current AI, on its own, is capable of this either. I had never worked on or thought too much about the Riemann Hypothesis: it seemed very far removed from my own area of number theory. But the astonishing thing about Levent's claimed proof of RH (which I think has a very good chance of being correct) is that it uses a huge number of precisely the things that are central to my own corner of number theory: - Theta functions and other automorphic forms on unitary groups - Gan–Gross–Prasad period formulas - Relative trace formulas Etc., etc. A problem that I had always assumed would require new theories in arithmetic geometry may instead turn out to be tractable through an inhuman chain of existing (deep) results in higher-rank automorphic forms (and a lot else).
This actually makes me a bit more confident that the proof has a good chance. RH is proved by using *everything*!
154
667
145
6,704
697,459
Just horrible. Eva is an insightful and kind researcher who I’ve known personally for several years. Hard to understand the function of harassing traveling researchers beyond inducing terror & chilling speech.
Our colleague Eva Yi Xie, a Princeton neuroscience PhD student, was detained by ICE at Reagan National Airport on her way home from SfN, where she’d just presented her poster. She’s being held in Chantilly, VA. Please help us bring her home. 📷
12
5
67
2,268
The most ambitious, competent, humble, and joyful people I've ever worked with. Thank you.
48
721
Applied Compute is nothing without its people 🧡
48
47
7
1,401
85,864
He's a good kid @ypatil125
So proud of you, little bro. A lot to celebrate this Thanksgiving.
17
26
4
1,289
162,531
There are millions of reasons for them to stay for the next 3-4 years, they will retain their offices and independence for the most part.
This is a phenomenal acquisition for $PANW and a great tie up for Applied Compute which is increasingly in the compute business. It will be interesting to see what talent retention looks like. PANW is an incredible company but by all accounts not a fun business to work for given the drudgery of low level network optimization eng and product work. Great engineers attrit quickly from past acquisitions, even w/ meaningful 3-5yr comp packages. My guess is that PANW will wait a while before deeply integrating this team into their broader org.
26
20
4
682
76,876
Niteshift has genuinely changed the way our team works. I haven't made PRs from local dev in weeks.
- Listen Labs starts using Niteshift [Jan 2026] - Listen Labs acquired for $2B [Sep 2026] - Applied Compute starts using Niteshift [Sep 2026] - Applied Compute acquired for $4.5B [Nov 2026] Sign-up here: niteshift.dev
5
1
20
2,536
We're just getting started!
Applied Compute has signed a definitive agreement to be acquired by Palo Alto Networks for $4.5B! Rhythm, Linden, and I founded Applied Compute with a belief that was unusual at the time: that every company should own its intelligence. We think training, serving, and improving proprietary models will become a normal part of running a company. To do useful work, a model needs to access sensitive data and systems, comply with policies, and remain explainable, verifiable, and auditable. This is the reason we've always prioritized security, and why joining Palo Alto Networks will help us go further faster. Customers will still own and control their models, and I'm excited for all the ways our combined capabilities can serve them even better. Above all, I'm incredibly grateful for the AC team. A few dozen people in just over 18 months grew this company from zero revenue to over $100M in annualized revenue run-rate, working with many of the world's most innovative companies. More to come soon.
4
33
913
SFT as a training technique is generally very underrated. People have gotten this idea that you can’t SFT without catastrophically degrading a model. This is completely untrue. Done well, you can often come close to RL performance with an order of magnitude less data (1/n)
20
36
4
810
49,091
Great work! This event was really fun, and it was exciting to see so much variance in the successful strategies :)
Hell yeah, we won the pockerbot track and $7500 at the @ValsAI × @appliedcompute hackathon 🏆♠️ In 24 hours, we post-trained Qwen3.6-35B-A3B on 10 B200s to play 6-max No-Limit Hold’em and it beat every other bot. Huge thanks to my teammates @notionsmith and @gautam_jajoo Thanks to @vinj_ai and @appliedcompute team for compute and support and to @RayanKrishnan for an amazing hackathon.
1
17
1,366
vinjai retweeted
Surprise! Vals AI is adding a third track backed by @appliedcompute, offering a $7,500 cash prize for the first place team. Alongside Build Your Own Eval and Post-Train a Small Model, a limited number of teams will post-train a model to compete on PokerBench, Vals’ 10-player no-limit Texas Hold’em table. Every team starts from the same reference setup and gets the exact harness their bot will be scored in. Bots play each other all weekend, with live rankings and overnight tournament checkpoints. First, second, and third place take home cash prizes.
1
6
4
46
18,757
Low-precision is the future. Frontier open models like 4.1 Flash and K3 already serve inference with low precision. I think the same thing will happen soon for training. Training actor forwards, backwards, and even optimizer state will move to 8-bit and 4-bit regimes.
1
6
6
48
18,757
AWS : CPU :: AC2 : GPU
AC2 = Applied Compute Agent Cloud
16
816
This is it. If I were in charge of a company, it'd be my number 1 priority to get this working ASAP. No RL envs, no simulated tasks, no large efforts for data creation. Just straight production traffic -> weights.
Production traffic is full of real experience from economically valuable work, but very little of it currently makes it back into training. We built a recipe for production RL that steadily improves a model straight from its deployment’s data stream, then redeploys it to keep learning. We are sharing some of our initial learnings from a recipe capable of continual learning in production.
13
10
1
271
26,509
You already know the dude holding the snorlax is cracked
Applied Compute update: - crossed ⁠$100M annualized revenue run rate up from zero 18 months ago - raised $350M at $3.25B led by Elad Gil - 40% of customers use AC for both custom model training + production inference - inference serving 1T tokens a day after launching 12 weeks ago - 2X enterprise customers since last raise - 3X team since January, 85% are researchers and engineers ⁠- Working with customers like Microsoft, Shopify, Harvey, DoorDash, Cognition, Clay, Decagon and others To move even faster we need people who are extremely technical AND extremely practical. If this is you, please reach out.
3
8
1
213
45,503
Strongest post training team💪🏾
Applied Compute update: - crossed ⁠$100M annualized revenue run rate up from zero 18 months ago - raised $350M at $3.25B led by Elad Gil - 40% of customers use AC for both custom model training + production inference - inference serving 1T tokens a day after launching 12 weeks ago - 2X enterprise customers since last raise - 3X team since January, 85% are researchers and engineers ⁠- Working with customers like Microsoft, Shopify, Harvey, DoorDash, Cognition, Clay, Decagon and others To move even faster we need people who are extremely technical AND extremely practical. If this is you, please reach out.
38
22
24
937
151,678
I built a bot that reviews code like me, then had Ari, our agent on AC2, autoresearch an eval for it and hill-climb it overnight. Bug bots like Codex are great for coverage, but they don’t know which shortcuts I’ll let slide, or that I review every PR asking “would we build it this way from scratch?” It just doesn’t feel like I’m getting a review from someone that understands our codebase or team. Code review has become Applied Compute’s biggest bottleneck as we write code faster. @felixs8696 joked I should build a bot since I was so behind on his PRs, so I did. I turned 3,500 of my comments across 2,000 PRs into an agent that talks and thinks like me. I’d like to think I am more complicated than an agent, and on the first pass it seemed I was right. The first version was noisy, leaving a pile of comments where I’d leave one or two - I couldn’t trust it. So I asked Ari, our agent on AC2, to build an eval and hill-climb it, prioritizing precision so the bot wouldn’t flood PRs with comments I’d never make. I turned on autoresearch overnight, and Ari: - Built a dataset of my past reviews, checking out each PR exactly as it was when I reviewed it - Built graders that verify every finding against the real code and my actual comments - Stress-tested those graders on 50 adversarial cases before trusting them - Ran ~100 experiments across prompts, models and harness changes, confirmed on a held-out set By morning, reviews were ~11 points more precise and used 22% fewer tokens. The biggest win came from how the bot explores the codebase, not from the prompt. Now it’s my first pass on my own PRs, and I trust it because it was measured on my own data. If you want to build your own evals on AC2, reach out! Checkout the autoresearch blog post here: appliedcompute.com/research/…
2
23
536
Today, Ari is getting a promotion from Applied Research Intern to Applied Research Intelligence. Awesome work by @liujaray @samueldenton @aniruddhsr24 and team.
Post-training a model usually takes a team of researchers running experiments for weeks. With Ari Autoresearch, now live on AC2, an agent can run that whole loop itself. In one 16 hour episode with no researcher intervention, it took Qwen3.5-9B from 0.118 to 0.629 held-out reward on chess puzzles.
1
31
1,032