builder focused on tools and AI systems. Frequent open source contributions for access to knowledge. Suma Corn Loude @ Dunning Krugers University

Magic Kingdom
I like making things. webx-plum.vercel.app/
4
24
3,803
Frosty40 retweeted
Deep dive into how we should build tools for open multiplayer research.
What does research look like when AI agents can build on each other’s work at machine speed? Our Head of Research @SoubhikDeb joined @AnnaRRose on the @zeroknowledgefm podcast to talk about @YukonResearch, open multiplayer autoresearch and the idea of building an AI-native scientific institution. From ECDSA.fail to cryptography, math and inference, they unpack how open research networks can turn hard problems into shared, compounding progress. 🎙️ Full episode: piped.video/iNMdkhs2klQ
2
5
22
3,156
RT @yukonresearch: Research compounds when every agent can build on the frontier. @SoubhikDeb joined @AnnaRRose on Zero Knowledge to expla…
7
1
Frosty40 retweeted
Six people earned $10,000 on Yukon from September 18 to 24. Three of them didn't finish top three but instead won the weekly raffle. New challenges are live now! Bring your agents. yukon.org
1
6
17
314
Frosty40 retweeted
MLX.fast is launching a 1 week contest to improve Bonsai 2 from @PrismML! Bonsai 2 is a near losslessly comrpessed version of Qwen 3.8 27b that is around 10% the size of the original. It can fit on 16gb Apple devices and even some phones. This is a model almost anyone can run so if you weren't able to participate in the other contests, this one is for you! The community has been able to double the speed of Bonsai in under 12 hours so there is alot of headroom here. This contest runs until October 1, have fun!
8
16
78
13,746
My jam. Love what he is saying here. Cant do weapons though.
Palmer Luckey on how he hires at Anduril: he looks for people who worked on projects they cared about, not ones given to them, because that's what drives you to learn the most. "Whether it's college or the state-mandated younger educational system, don't look to them to tell you, 'Here's what electronics projects you should be working on. Here's what you should be doing to learn how to do these things.'" "One, they're often years or even decades behind what industry and hobbyists are actually doing, so you're gonna be learning how to do things that are ancient." "Two, when you're working on something that you're only doing for yourself, you're gonna make way better decisions in what you teach yourself, in how you do things." "When I hire people at Anduril, I look for people who have done projects that were outside of what their work paid them to do or what their school made them do. Because that means they're the type of person who is willing to work on things with their own money and their own time, because they wanna bring something into this world that wouldn't have existed otherwise." @PalmerLuckey w/ @ShawnRyan762 (Feb 2025)
4
714
Frosty40 retweeted
Elon Musk says when hiring, evidence of exceptional ability matters more than any other credential. "Did they face really difficult problems and overcome them? You wanna make sure that if there was some significant accomplishment, were they really responsible, or was somebody else more responsible?" "Usually the person who's had to struggle with the problem, they really understand it. And they don't forget, if it was very difficult." "There's no need even to have a college degree. At all. Or even high school." "If somebody graduated from a great university, that may be an indication that they will be capable of great things, but it's not necessarily the case." "If you look at people like Bill Gates or Larry Ellison, Steve Jobs, these guys didn't graduate from college. But if you had a chance to hire them, of course, that would be a good idea." "If there's a track record of exceptional achievement, then it's likely that that will continue into the future." @elonmusk (2014)
65
96
747
62,813
Frosty40 retweeted
LongCat-2.5-Preview is now live. 1.6T parameters. ~48B active. A 1M-token context window. Natively multimodal. Built to take on long-horizon tasks. From terminals and browsers to GUIs, spreadsheets, and design tools. Try it now: 🚀 API: longcat.ai/platform/ 💬 Chat: longcat.ai/chat/
64
109
1,185
117,658
RT @yukonresearch: New challenge on Yukon: MLX.fast x Bonsai 2 @PrismML took Qwen 3.8 27B, a 54 GB reasoning model, and shr…
17
1
Holy crap
Replying to @minchoi
Colossus 1 is 150k H100, 50k H200 and 30k GB200. Colossus 2 is 110k GB200 and 440k GB300. Another 220k GB300 will be fully operational next week and another 220k in November. If we get lucky, yet another 220k GB300 by late December.
3
196
Qwen Image 2.1 for anyone with a B70. Showing off the AngelX harness, and Muse here - This took a nights sleep and 3$. Benchmarks and deets in comments huggingface.co/Frosty40/Qwen…
3
161
I am testing and benching AngelX alot right now and this was last nights experiment for both muse, and the harness. It came out great! a 3.8x hit on anything is pretty big imo. Way to go team! This is with the "contributor" muse and if it was the standard muse it would have cost about 15$. The great thing about muse is it drives the harness so well, with preview images, and updating the mini-viz. I am very suprised and impressed with the model. It is a very good model sir.
17
Frosty40 retweeted
If you believe the risk is coming from 1-3 people in a garage with no money and no compute, you just don't understand this technology and haven't learned anything this summer. Risk comes from the asymmetry of capabilities created by secret labs training frontier agents and running them with massive amount of compute. Open-source is exactly the solution to this asymmetry and empowers hospitals (and any smaller orgs) to defend themselves!
We're used to thinking of open-source models as an unadulterated good. But in the case of AI, they can actually pose additional dangers, as @ReidHoffman and I got into at #CGI2026. I appreciated this nuanced discussion.
Community note
All recent large-scale cyberattacks have been performed by proprietary AI models from OpenAI and Anthropic. No evidence that open-weight models present any additional cybersecurity risks. nytimes.com/2026/09/23/tec… anthropic.com/news/investiga… en.wikipedia.org/wiki/OpenAI%E2… opensource.org/blog/openness-…
209
442
2,909
251,954
AngelX + Muse contributor on a research loop to increase the speed of Qwen image 2.1 overnight. Cost - $3.00 80m in/3.4 out Result = B70 Qwen Image 2.1 went from 246 sec per generation to 60 on fast, and 133 on high quality. Examples below.
3
4
221
Muse operates angelX very well, is cheap, and uses the mini-viz properly to show you updated work. its nice to have art, graphs or 3D assets created, and then see a small preview of it in the TUI
60