AI research. @CUBoulder, tiny-vllm, @PyTorch. Art by https://xitter.cf/t.co/rs0nMCNnN6

🇰🇷🇵🇱 Korea of Europe
Joined May 2023
Uncover the hidden truth behind softmax functions with @jedmaczan 🔢 👀 Watch the full recording to dive into the computational insights:  Thanks to our ML Math program leads @KatrinaL2899 and @neel_ghoshal for hosting this session!
6
1,792
I'm posting a series of prediction questions from Stanford CS312 (Deep Learning Alchemy) exams to test deep learning intuition. First up: optimization. If we increase an LM's batch size 8× (same token budget), can rescaling learning rate or weight decay compensate? Try below ↓ This quiz is part of our optimizer hyperparameter assignment and unit designed by @wen_kaiyue Setup: We train an 8-layer, ~35M-parameter Transformer on 614.4M tokens using AdamW. At batch size 64, the best reference run has peak LR 0.003, weight decay (WD) 0.2, and final validation loss 2.9178. Now increase the batch size from 64 to 512, which gives 8× fewer optimizer steps. Everything else stays fixed except the changes below. Which recipe gives the lowest final validation loss? A. Keep LR and WD unchanged B. LR ×√8 C. LR ×8 D. WD ×8 E. LR ×√8 and WD ×√8 What would you predict, and why?
13
63
9
685
55,783
Jędrzej Maczan retweeted
폴란드인을 한국인에 빗대어 열등감에 사로잡힌 민족이라 조롱하기 전에, 당신의 나라가 폴란드에게 무슨 짓을 저질러 버렸는지부터 제대로 기억해야겠지? 나치 독일은 폴란드를 침략했고, 수백만 명의 폴란드 시민이 독일의 점령과 학살로 희생됐어. 바르샤바는 폐허가 됐고, 폴란드 민족과 문화마저 말살당할 위기에 놓였었지.. 그런데 이제 와서 폴란드인들이 역사를 기억하고 안보를 걱정하는 것을 열등감이나 피해의식으로 감히 치부해!? 역사를 기억하는 것과 과거에 갇혀 사는 것은 전혀 다른 얘기야! 침략당한 나라가 다시는 같은 비극을 겪지 않으려 경계하는 것은 열등감이 아니라 역사에서 배운 교훈이다! 나는 한국인으로서 폴란드인들의 감정을 이해해. 국권을 빼앗기고 식민 지배를 겪었으며, 자유와 주권이 얼마나 소중한지 역사를 통해 배웠기 때문이야. 당신의 조상이 저지른 만행을 기억하는 사람들을 조롱하기 전에, 먼저 그 역사를 직시부터 해라. 역사의 피해자를 비웃는다고 가해의 역사가 희석되는 건 아니니까.. 🇰🇷🤝🇵🇱
Japaner haben das gleiche Problem mit den Koreaner wie wir das mit den Polacken haben. Sie haben auch einen Minderheitskomplex und jammern über den 2. Weltkrieg Reparationen und über den bösen Nationalistischen Nachbarn, der sie jeden Augenblick angreifen wird Finde diese Parallelen sehr interessant, dass 2 so weit entfernte Völker so ähnlich sein können. Japaner sind die Deutschen des Ostens und Koreaner die Polacken.
99
321
14
4,995
125,594
Jędrzej Maczan retweeted
Surprise 1: Anthropic acquired a one-person biotech . . . Surprise 2: that person is me Surprise 3: I’m here to tell you all about it Yes, this is what you think it is; after 6 yrs of increasingly loose anonymity: the Professor Oak reveal post …and yes, it has gotten so boring around here that I’m literally doxxing myself to spice up the timeline ------- Ok, allow me to introduce myself. My name is Cody Tranbarger. I mostly tweet, but occasionally find time to dabble in drug dev & starting/building biotechs. My past lives span the VC, crossover, startup continuum; Prof Oak has been my way of keeping a digital foot in those worlds, despite having my physical feet planted elsewhere. The past 18 mo, ‘elsewhere’ was Mithril Medicine, a co. I founded to develop bsAbs for LAMA2-CMD & other ECM-related genetic diseases. Mithril was born out of hands-on building with @MikeNGladstone, @JasonPaulRhodes, @LifeSciVC, & team @atlasventure. Somehow, I made enough half-decent calls along the way (even a broken clock is right twice a day, eh?) to convince them Mithril/I warranted a small seed. While I can hardly rationalize such an outrageously cavalier bet, I’m grateful for it, & I'd partner with Atlas again in a heartbeat. In typical AV fashion, Mithril’s primordial form was virtual, decentralized, & ultra-lean. This stage tends to be short-lived, but for Mithril, it worked surprisingly well, for surprisingly long, thanks in part to the rapid maturation of the AI tools I used liberally. Eventually, Mithril became a bit of an experiment itself - to find the upper bound of how far it could scale as a single operator + a bunch of Claude (+ a stellar bench of part-time consultants). In parallel, I started spending time with @nc_frey, @an1lam, & @CoefficientBio team; and promptly experienced, for the 1st time, AI at the frontier. Soon thereafter, Mithril stopped looking like an experiment, & started looking like a preview of an entirely new operating model. From that spark came “Pipeline-in-a-Person.” The idea immediately resonated when we shared it here back in April; and it was central to the shared vision that led to Anthropic’s acquisition of Mithril, & my joining the team. Why join Anthropic? Three reasons. (1) For Mithril, the programs continue, with more resources & infrastructure behind them, and as much Claude as I can put to work. (2) For me, the chance to start several more programs, & find out how many one person can actually run in parallel - an almost unprecedented luxury for a curious drug hunter. (3) For our mission, a path to institutionalize & scale this new way of working, then put it in the hands of all of biopharma, all over the world - which, candidly, I find the most energizing. Will AI x Bio cure diseases? Unequivocally yes, imo… but not by the route that gets the most airtime. Rather than timeline acceleration & compression, I see orchestration & parallelization as today's low-hanging fruit. Most drug hunters see a handful of go/no-go calls in a career, & spend entirely too much of the rest trading emails (my inbox can attest). Give them the bandwidth to run several programs at once, and the time to focus on scientific judgment, and I’m willing to bet they’ll make better calls. And at scale, better calls translate to more drugs, higher hit rates, broader patient impact, & likely, a disease or 10 or 100 cured. I couldn't be more excited to build toward that future @AnthropicAI. Finally, a heartfelt thank you to BioTwitter/X. The end of anonymity is bittersweet, but if anyone’s earned a proper introduction, you all have. It’s difficult to overstate how much this app has done for me: a decade of education, plenty of friends, many job opportunities, several mentors, a few investors, even a co-founder (fmr). It was the generosity of this community that inspired me to start posting 6 years ago, after just as many years lurking. I hope I’ve been able to pay it forward, in my own small way. But, to be clear, Prof Oak isn’t going anywhere. Same beat, same themes, and now, a front-row seat to drug development at the frontier. Stay tuned for dispatches from the trenches; and remember... Gotta Cure 'Em All
448
819
294
11,641
1,128,618
Jędrzej Maczan retweeted
I think it's becoming rather clear that ALL of mathematics will become possible to formalize rather soon. There are deep implications: 1. Mathematics becomes super interactive. Formalizations make it easy to browse, recombine, and build mathematical proofs from components. 2. Each new idea and paper can be checked much faster. The deep part is meaning and explanation. Validation is almost automatic. 3. Optimization and identification of new proofs can be done partially via agents, and humans can contribute much faster with great new ideas. Proofs from the Book are possible at scale. 4. Scientists and people at large can finally use mathematical artifacts as black boxes with trust and confidence. We can commoditize mathematical theorems. I think the mathematical enterprise also becomes very gamified. We can clearly see each unproven statement as a concrete goal, identify existing tools and directions, and focus quickly on what's missing. And everyone can contribute, with agents or through understanding. Soon, new languages and new theories will also develop, but the structure will keep growing steadily, with no mistakes and on fundamentally solid foundations. Hilbert's dream is becoming a reality. One important caveat: mathematics is by no means trivial now! It's just becoming extremely well-structured and laser-focused. I think the jump in capacity, quality, and scale will be massive. I also hope that our understanding of the big picture will remain solid, because there will be a lot to grasp.
15
31
4
250
12,483
Jędrzej Maczan retweeted
A potentially controversial thought: We should make AI PhD coursework only optional. So many of our PhDs are tied in numerous classes, but: - Self-learning is easier than ever. - Many incoming PhDs are already strong researchers. - We need to free Profs' time for more research mentorship. - Urgent AI problems need free-thinking minds (like our academic students!), independent of industry for-profit directions. Courses are valuable. But they don't need to be mandatory, especially when so many problems with AI are threatening our society, and urgently need to be solved.
34
21
4
303
27,099
Jędrzej Maczan retweeted
Kind of crazy that the same week arXiv announced a 2-paper submission quota, OpenAI dropped hundreds of preprints on GitHub
2
21
640
Raising public awareness on cybersecurity misogyny
41
42
7
397
24,505
Jędrzej Maczan retweeted
🇵🇱 Okay my final Poland video because we're leaving soon! It was really fun and we had a great time I liked Warsaw most to be honest, a very modern city, a bit like East Asia in terms of modernity, with rapid delivery services and apps, everything just works great and fast I'm not saying it's perfect btw, there's moments where you still realize it's Eastern Europe (or as they'd like to call it Central Europe) but then the difference is they generally solve problems fast and pragmatically Many shops you go will actually have a long line at the checkout, annoying, but usually there's a second checkout but it's closed That's because Poland too has a big labor shortage, one of the lowest fertility rates in Europe, so they too kinda have to import people, but they choose to import them from Ukraine and Belarus (and some South Asians actually too) But wait, so that checkout line is closed yes? I'm so used to Portugal where there is zero incentive to then open another checkout counter but in Poland usually within a few minutes they notice it's busy and open another counter. I know, basic stuff but hard to come by these days. They're fast and pragmatic Ordering flowers for my gf was easy, the flower shop has an IG (but I don't) so I msg them over Whatsapp, they immediately respond, give me some options, I say what I want, address, payment link with Stripe (!) and done delivered at my hotel by a nice guy. All soooo smooth. Fast response, friendly people. Things just work here! Poland has done a great job keeping bad immigrants out while bring good immigrants in. The Ukrainians and Belarussians are neighbors and their cultures overlap strongly with Poland. So they fit in easily. Same cultures, same religions, same routines etc. for the most part A young Polish Uber driver I talked to said to him Krakow (his city) looks just like Lviv in Ukraine, so it's easy for Ukranians in Poland to integrate There's a lot of police presence in places, and military vehicles also (probably cause of the Ukraine war), the security at the Warsaw Airport is not done by some 5 layer outsourced company with staff that hates you, but by friendly people from the Polish Border Guard (Straż Graniczna) wearing military uniforms Warsaw has lots of skyscrapers now, and actually the tallest in Europe now at 310 meters is the Varso Tower, crazy! Poland is a booming economy and growing fast. When you're growing of course it's easy to make things go well. I'd say it's important for Poland to look ahead carefully and don't make the same mistakes that Western Europe did when everything was going well and we were very rich We became complacent, stagnant, added regulatory burdens, became overly obsessed with climate change and the environment, and imported the lowest quality tier of people from cultures that hate our guys and are diametrically opposed to Western culture That last part is especially interesting considering Poland's history, where 6 million (!) or 20% of the country was ruthlessly murdered in World War II I think that gives them the power to not have to be politically correct about immigration like Germany, because Poland after all was the victim, while Germany then was the purpetreyor. Which is why Germany does have to be so politically correct about it, due to extreme historical guilt You walk on the streets of Poland in the day and at night and you see the difference what sane policy does to daily life We're all dealing with the echoes of history So my message to my Polish friends: don't make the same mistakes Western Europe did and you will soon be the richest, safest, most modern country in Europe and hopefully then have the power to be the new leader of Europe And maybe you can then save Europe as a whole!! Niech żyje Polska! 🇵🇱
🇵🇱 Warsaw is Poland's biggest city and it is completely safe at 1am at night No violent junkies, no drug dealers, no street thugs hanging around harassing women (and men), no pickpockets (nor in the day btw) No shady people hanging around the train and metro stations Just people living chill together and police enforcing safety You know like we used to in Western Europe? There's a lot to learn from Poland
180
287
31
3,755
215,938
Jędrzej Maczan retweeted
2023: GPT-4 almost solved grade-school math 2024: IMO silver 2025: IMO gold 2026 May: Erdős unit-distance conjecture 2026 Sep: Navier–Stokes Millennium Prize Problem solved 2026 Oct: 722 AI-written math manuscripts Today: OpenAI proves ETH is false Tomorrow: P = NP?
122
210
34
4,088
217,554
Jędrzej Maczan retweeted
As was well-known, and circulated a number of times even on here, Ethan Sussman has been writing up existence of YM4 publicly via Balaban’s approach, and has posted existence of QED3 and YM3 and an old preliminary outline of YM4 on his repo, working with countless community members in understanding the arguments. I kept telling him to not be too deferential and modest because he would be publicly scooped at any opportunity. Instead of contacting him to work with him, as has been suggested publicly many times, the people at OpenAI chose to do this. “Our work builds on Balaban’s approach⁠ in constructive quantum field theory. As a subproblem, this work also includes a proof of the existence of pure Yang–Mills theory, which is part of the Quantum Yang Mills Theory Clay Mathematics Institute Millennium Prize problem⁠.”
50
65
21
804
112,173
Jędrzej Maczan retweeted
I've noticed something puzzling and exciting when I work with abstract algebra and models. In physics, we have the notion of supercriticality, i.e., when you follow the liquid-gas coexistence curve as temperature and pressure increase, you reach a critical point beyond which matter enters this weird state where it is neither a gas nor a liquid. The distinction between the phases disappears, and it's hard to tell what you're looking at. I want to say that I've recently experienced this phenomenon a lot in the AIxMaths space, in the region where we consider highly abstract notions from universal algebra. At some point, the level of abstraction becomes so high that you end up talking about objects that are not intuitive to people at all. Every insight in this space comes basically from text, and no human can claim to be superior to LLMs in terms of originality, creativity, etc. Because everything at this level of abstraction is based on text. The new notions, ideas, and progress basically come from text processing, and to admire new theorems, you have to get used to them. This is something I see personally when I work with semigroups, loops, and other arcane objects using models. One can prompt agents to make synthetic discoveries, and they are capable of original ideas, new insights, etc., because all of this is based on pure text. Even the legendary Irving Kaplansky, when asked what he pictured when thinking about algebraic rings, reportedly said, "I see the letter R." And I think this is much deeper than one might think. Because at some level of abstraction, we enter a supercritical zone of symbolic thinking, and at this level none of our intuitions has a natural background. Every insight is superhuman, alien, and artificial. It's a no-man's land, and everything there is hard-earned. And LLMs are fast and capable. I'm learning now that, once guided well, they can make artificial discoveries. More to report soon. Stay tuned!
17
7
3
90
7,502
An internal model at OpenAI has produced a claimed refutation of ETH! Russell Impagliazzo, Mohan Paturi, and I have been asked to help organize an open and independent examination of this claim. 1/4
293
746
530
7,262
3,597,009
Jędrzej Maczan retweeted
Can someone reach out to me in dm so I can rent for a few days bare metal Mi350/MI355X? I need near bare metal, not fully virtualized or at least allow me to customize hw virtualization. @tensorwave @CoreWeave @AIatAMD 🙏
3
1
1
6
1,489
What's happening out there? My preprint which is already published in IEEE journal, has a DOI (which I provided with the submission) got rejected from @arxiv??
1
102
Twitter is really pay to win now.
1
2
230
Jędrzej Maczan retweeted
This is the most valuable free resource we've created on AI Evals 🎉 (not exaggerating!) I organized our public materials into this guide. It allows you to find answers to your eval problems w/o searching aimlessly. Humans: pick the row that sounds like you. Agents: point it at the post The material draw on 60+ hours of office hours from our Evals course, where @sh_reya and I have taught 5k engineers and PMs Evals. We add new material often, with 15 FAQs added in the last two weeks. Recent additions are marked with a "New" badge. You can find all of these and more here: hamel.dev/blog/posts/evals-f…
69
130
16
1,207
130,414
Jędrzej Maczan retweeted
If you've ever wondered why smart TVs are so cheap, this is why. @log4j rooted his Samsung and found an always-on hot mic, automatic content recognition, geolocation, network mapping, ad targeting and user profiling, shared with Samsung's "400 partners". A lot of it can be blocked in <15 min, wrote a basic guide below. What your TV is doing The main culprit is ACR (automatic content recognition). Your TV screenshots what's on screen, fingerprints it, and sends it home to build your ad profile. Samsung does it every 500ms, LG every 10ms [1]. What is insane: it happens on connected devices too. Plug in your laptop, your console, your work machine, and your TV can fingerprint that via HDMI [1]. He also found that if your Samsung doesn't have a mic, it'll look for one from a connected soundbar or bluetooth device. xitter.cf/log4j/status/211376224… It's on by default, and you likely agreed to it on the setup screen you clicked through in 10 seconds. Step 1: Turn off ACR Samsung: Terms & Privacy → Viewing Information Services LG: Live Plus Vizio: Viewing Data Roku: "Use info from TV inputs" Sony/Google TV: Samba Interactive TV Menus vary by model, but those are the names to look for. While you're in there, opt out of interest-based ads, reset your advertising ID, decline every optional privacy agreement, turn off voice recognition, and delete apps you never open. Step 2: Give smart devices their own network Put the TV, cameras, speakers and vacuum on your router's guest/IoT network. The logic is simple: these devices rarely get patched. If one gets compromised (getting drastically more frequent with AI), it shouldn't be able to see your other devices or home network. Step 3: Block trackers for the whole house with DNS-level filtering NextDNS (cloud-based) is easy and has a solid free plan. Pi-hole is a little bit more work if you prefer self hosting. For simplicity we'll focus on the former. 1. Sign up at nextdns.io (free tier is plenty) 2. Privacy tab: add a blocklist (OISD is a good bet) and turn on Native Tracking Protection for Samsung, LG, Roku, Amazon (see attached image) 3. Setup tab: copy your endpoint 4. Paste it into your router's DNS settings as DNS-over-HTTPS/TLS (or use the IPs + "Link IP" if your router only does plain DNS) 5. Bonus: block outbound port 53 except to your router. Some devices hardcode their own DNS to get around you Then open the NextDNS logs and watch how often your TV phones home. It's eye-opening. If an app or update breaks, find the domain in the logs and allowlist it. Why they do this TV prices are down 97% since 2000 [2]. Most of that is better manufacturing and brutal competition. But once margins on the hardware hit zero, the cost moved to you. In Q3 2024, Vizio lost $6.7M selling TVs and made $115.8M in gross profit from ads and viewing data [3]. They've said on earnings calls that they keep TV prices low to grow the audience they sell to advertisers. Walmart bought them in 2024. This isn't a conspiracy theory either. Vizio paid $2.2M to the FTC in 2017 for tracking 11 million TVs without consent. Last December, Texas sued Samsung, LG, Sony, Hisense and TCL over ACR. Samsung and LG have since agreed to stop collecting that data without explicit consent [4]. TL;DR 1. ACR off 2. Ads + voice off 3. Smart devices on their own network 4. NextDNS filtering on the router Sources: [1] ACR study: arxiv.org/abs/2409.06203 [2] Cost of US consumer goods: aei.org/economics/chart-of-t… [3] Vizio Q3 2024: sec.gov/Archives/edgar/data/… [4] Texas v. Samsung: texasattorneygeneral.gov/new…
i own a relatively new (2025) samsung tv. i kept getting weirdly targeted ads. i tried reversing the firmware but it was aes encrypted and no keys unless tear it apart and do uart/jtag magic. zero days it is. root shell on new samsung TVs! discovered PRIVACY HORROR. write up?
351
3,274
339
22,561
1,507,946
Jędrzej Maczan retweeted
How should we approach research, when the very field we work on is going through rapid and potentially fundamental changes? I recently gave a talk at the Stanford Vision Seminar, reflecting on my PhD journey and how AI is changing the way I think about research. I adapted part of it into a new blog post, organized around three questions: 1. What should we work on? 2. How should we work with AI? 3. What is a PhD for? I hope some of my reflections are helpful. The talk ended with an open discussion, and I’d love to continue it here; curious how others are thinking about these questions. Personally -- as much as there's FOMO, I also think life is very long, and we have time. toruowo.github.io/blog/posts…
12
85
15
578
63,421