Vinay Ramasesh
@vinayramasesh
Followers
2K
Following
2K
Media
7
Statuses
445
Research scientist @DeepMind working towards a better understanding of deep learning. Physics PhD @UCBerkeley
San Francisco, CA
Joined April 2017
RIP my great friend Daniel Naroditsky , I still can’t believe
124
314
12K
@emollick i've spent serious efforts working on developing math problems with an unambiguous (e.g., numerical) answer that gpt-5-pro cannot solve. it is *nontrivial* to do so. it was totally different even 4-6 months ago.
12
31
286
Love to see @UCBerkeley & @UCSantaBarbara scientists win the Nobel Prize in Physics! Hate to see the White House celebrating breakthroughs from universities they’re trying to defund.
Congratulations to John Clarke, Michel Devoret, and John Martinis for winning the @NobelPrize in Physics! Their brilliant work at UC Berkeley demonstrating quantum tunneling in electric circuits has been critical to our understanding of quantum computing.
173
815
5K
Today, @ekindogus and I are excited to introduce @periodiclabs. Our goal is to create an AI scientist. Science works by conjecturing how the world might be, running experiments, and learning from the results. Intelligence is necessary, but not sufficient. New knowledge is
432
451
4K
Interesting post by Scott Aaronson describing a recent use of GPT-5 in his own research in quantum complexity theory:
scottaaronson.blog
Update (Sep. 29): Since this post has now gone semi-viral on X, Hacker News, etc., with people arguing about how trivial or nontrivial was GPT5’s “discovery,” it seems worthwhile …
0
1
5
Highly recommend this update from Jacob and coauthors:
Today we're putting out an update to the JAX TPU book, this time on GPUs. How do GPUs work, especially compared to TPUs? How are they networked? And how does this affect LLM training? 1/n
0
0
15
Interesting and seems basically correct
Multiple AI systems won gold medals at the 2025 International Mathematical Olympiad (IMO). Exciting as that sounds, @GregHBurnham argues that it represents little progress: an unlucky draw of problems made the event relatively uninformative. Is that cope? Judge for yourself. 🧵
1
0
12
maybe a better headline would be that oai and gdm ranked 27 at the IMO. some talented kids here!
6
13
263
This is the most scaling-pilled project I've ever been part of, and the team really cooked. TL;DR: With RL and inference scaling, Gemini perfectly solved 5 out of 6 problems, reaching a gold medal in IMO '25, all within the time constraints of 4.5hr.
An advanced version of Gemini with Deep Think has officially achieved gold medal-level performance at the International Mathematical Olympiad. 🥇 It solved 5️⃣ out of 6️⃣ exceptionally difficult problems, involving algebra, combinatorics, geometry and number theory. Here’s how 🧵
17
28
534
An advanced version of Gemini with Deep Think has officially achieved gold medal-level performance at the International Mathematical Olympiad. 🥇 It solved 5️⃣ out of 6️⃣ exceptionally difficult problems, involving algebra, combinatorics, geometry and number theory. Here’s how 🧵
157
742
4K
“Where did you meet the bootleg peptide dealer?” “At the underground robot boxing ring” Visceral future-shock
12
31
509
We desperately need to take all the people in SF who talk about accelerating biology but have only ever done math or CS and have them do a six month rotation in a wet lab.
“Finally, maybe this is controversial but ultimately progress in science is bottlenecked by real-world experiments.” If this is controversial in SF, we’re cooked.
28
48
678
Very excited to share this resource that @AdamMarblestone and I have been working on for the past few months! It’s an interactive map of scientific problems and potential solutions. What’s most exciting? What’s missing? Let us know! 👇
we made a map! https://t.co/YtwACsfSiP is a tool we built to help you explore the landscape of R&D gaps holding back science - and the bridge-scale fundamental development efforts that might allow humanity to solve them, across almost two dozen fields
2
7
26
LLM evals have the glaring blind spot of being focused on "things AI researchers understand". I appreciate people like Adam using their expertise to help, like in this post on evaluating how useful LLMs are at instructing on how to make metal parts. Spoiler: They're terrible
New blog post! I conduct a detailed case study on manufacturing this simple brass part with frontier AI models. Spoiler: All models except Gemini 2.5 have horrible visual abilities, and all models fail at the physical reasoning tasks. I speculate on what this means. 🧵
7
10
208
@ArmandDoma There's just no evidence for the lab leak, and quite a lot for zoonosis. (I used to engineer viral vectors for a living.) The rootclaim debate with Peter Miller and Saar Wilf is long but the best public airing of the evidence.
youtube.com
6
1
4
We can use a harmless wavelength of light to kill nearly all airborne pathogens. The technology exists! It works! And now there is a plan for bringing it out into daily life.
Far UVC can cut airborne bacteria by 98.4 percent, and could do the same for viruses, preventing diseases spread in public spaces. But it is held back because it is unpatentable, which means it is unproven, unregulated, and untrusted. We can fix this.
16
44
452
In their roadmap, Microsoft described a protocol for demonstrating a topologically protected qubit. There is no publicly available evidence that this test has been conducted successfully. I hope we will hear more soon. https://t.co/G97mdJnJGD
arxiv.org
We describe a concrete device roadmap towards a fault-tolerant quantum computing architecture based on noise-resilient, topologically protected Majorana-based qubits. Our roadmap encompasses four...
11
86
466