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)
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
15
810
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
45
10,200
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
5
6
41
7,920
AWS : CPU :: AC2 : GPU
AC2 = Applied Compute Agent Cloud
16
705
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
268
26,287
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
214
45,440
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
21
23
936
150,837
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
461
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
999
vinjai retweeted
SITUATION EXPLAINED: Applied Compute, started by 3 ex-OpenAI researchers, hit a $100M revenue run rate in under 18 months. • It went from a $20M seed in June 2025 to an $80M round last October to this $350M raise • Gil also backed the $80M round, and NVIDIA, Blackstone, Kleiner Perkins and Lux joined this one • Co-founders Yash Patil, Rhythm Garg and Linden Li are 23, 24 and 25 and became friends at Stanford • At OpenAI, Patil worked on what became Codex, Garg on reasoning models and Li on machine learning systems • It builds custom models on top of open-weight models and says they can be about 10x cheaper than frontier models in some cases • A custom model for Harvey improved answer and citation quality at under half the inference cost of a frontier model • Enterprise customers have more than doubled since the last raise, and its inference platform now serves over 1T tokens a day @sophiadew: "The big prediction that Elad had is that open models didn't have to defeat OpenAI for a market to become enormous. You're going to see the frontier labs dominate, but you're also going to see that wider ecosystem develop as well."
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.
5
6
54
10,711
Love Applied Compute's vision of owning your own intelligence. It's been awesome working with their team as a customers of niteshift.dev, and testing out their new Shift-1 model router (now supported in Niteshift!)
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.
1
4
2
37
2,952
Avengers... assemble!
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.
2
29
955
vinjai retweeted
harness matters! droid 1.0 is faster and more cost effective than ever. excited to reveal how droid works under the hood with programmatic tool calling - turns out models just want to write code been fun working with @luke_alvoeiro and @ain3sh on this!
Replying to @factory
On Terminal-Bench 4, Droid outperforms popular models’ native harnesses, with up to 20% better efficiency.
3
2
26
697
Really awesome research from the team: appliedcompute.com/research/… We're super bullish on routing as the first step to cutting costs and optimizing workloads. From what we've observed, routing policies will differ by domain (and the family of models you route across). This is super interesting because it supports the thesis that intelligence is inherently jagged and there is a best model for a particular job.
On Wednesday, we launched Shift-1: a model router that can reduce coding agent spend by up to 55% while preserving performance. Today we are sharing more details about the research that went into this release.
1
1
48
3,711
The point about employee discontent is quite salient. From the perspective of safety activism outside OpenAI, neither OAI's revenue nor the government are currently reliable/actionable levers to incentivize them to adopt safety measures -- employee pressure is the primary one right now. It's also most likely not the case that OAI leadership is somehow nefariously plotting to undermine safety. They are simply following the path of least resistance to appease their board and their own conscience.
I'm just speculating like everyone else (no juicy secrets about this shared with me) but here's my guess about what is going on right now with OpenAI and the three fired employees: The negotiations between OpenAI and METR are ongoing. OpenAI has said they want embedded third parties, METR has proven their worth with the previous engagements, so METR is an obvious choice. But my guess is that METR is asking for more this time; probably they want more access, more time on site, less narrow scope, the right to comment on redactions, etc. (Compare with the terms of their previous engagements: In this one teapawt.org/METR_Evals/status/2113… a third of their report was redacted and they weren’t allowed to say why, and in the Hugging Face investigation they only had 6 days and were only allowed to investigate the precursor swarms nor the subsequent more severe compromise of OpenAI infrastructure. teapawt.org/DKokotajlo/status/2092…) Various researchers at OpenAI probably support METR in this (including, I’d imagine, the fired three employees? Tomek was the point person on the hugging face investigation), but also various higher-ups at OpenAI and legal types probably oppose and want more restricted access and scope – basically they want to work with an org that’ll go easy on them and not give them trouble. OpenAI holds most of the cards here, since this whole thing is voluntary and not legally required, and they can threaten to go with Accenture or HuggingFace or whatever instead. Anyhow to improve their bargaining position and/or minimize the blowback if they decide to say no to METR and go with someone more pliable, it helps if they can discredit METR. Hence (I suspect) why they are going after my friend Ryan Greenblatt (who works at METR) now. They don't need to present actual evidence, they just need to cast vague aspersions, and many employees will give them the benefit of the doubt. If and when they say "We chose to work with a different third party" they'll face lower risk of employee discontent than if METR's reputation was unsullied. This whole mess, by the way, was predictable. *Of course* there's a big political fight about who should be trusted to do this and who has relevant expertise and independence etc. and *of course* the companies are trying to make things shake out in a way that favors them. It's part of why I've been advocating for *actually pacing the frontier* instead of just doing third party assessments. "We allocate at most 20% of our compute to AI R&D" is a lot easier to verify than "we are being safe and responsible" and depends a lot less on subjective judgment and technical alignment expertise -- AND, if you do it, people will be able to see the slowdown in the trends of agentic coding performance in the frontier companies and see laggard companies start to catch up. Harder for it to end up just being safetywashing / industry capture.
2
9
1,501
Model routing is a interesting technical problem. It's quite hard to build an all-purpose general model router that transfers to new settings. AC2 Routing flips this problem on its head, because we make it accessible for anyone to train a router on their specific distribution. Then "overfitting" is just "fitting". Shift-1 is the result of us fitting a model to our coding agent usage and we saw some pretty incredible savings (55% in Codex) with zero observed intelligence degradation across the company. Let us know if you want the same results for your team!
Introducing Shift-1, a frontier coding router trained on AC2 that reduced coding agent cost by up to 55% with no noticeable drop in quality. Try it at shift.ac.
2
22
585
Shift-1 is available in a free research preview, powered by our new Routing infrastructure in AC2. Routers are a composable primitive for deploying flexible inference workloads. With AC2, you can ship smart auto-routing with custom decision models, autoscaling, A/B tests and shadowing, online evals, and more. They are a critical piece bridging between the model factory and prod. Routers are available starting today for our customers to use in AC2. Reach out if you are interested in building your own routers for your workflow!
Introducing Shift-1, a frontier coding router trained on AC2 that reduced coding agent cost by up to 55% with no noticeable drop in quality. Try it at shift.ac.
2
25
917
is your agent swarm really Recursive Super Intelligence or is it just Runaway Spend Inflation
9
372