←
Anton Tsitsulin1,880 posts
Anton Tsitsulin avatar

Anton Tsitsulin

@graph_

Gemini and Gemma, mostly pre-training (又菜又爱玩). past (?) life: graph machine learning

◷ Joined August 2016▧ 258 Photos and videos
Omar Sanseviero avatar
⟳ Anton Tsitsulin retweeted
Omar Sanseviero@osanseviero·Nov 17

With Carbon's help, I ran Gemma directly in an office printer 😅🖨️ 🤏Carbon compiled a small transformer engine + quantized Gemma 3 270m into a single PostScript job ⚡️ 100% in the printer: 14MB peak RAM and blazing 60s per token 🧠Smart layers juggling to avoid OOMing

Post attachment
Ethan Epperly avatar
⟳ Anton Tsitsulin retweeted
Ethan Epperly@ethanepperly·Nov 3

Delighted to share that @_kreda_, Yuji Nakatsukasa, and I have written a survey paper on randomized linear algebra. We tried to fill each page with our excitement for all that our research community has accomplished. Take a look! arxiv.org/abs/2610.38157

Post attachment
Anton Tsitsulin avatar
Anton Tsitsulin@graph_·Oct 31

Pope as an invited speaker at reliablemlworkshop.github.io

The future of #ArtificialIntelligence is not pre-determined. As a technology developed by human beings, its future depends on the choices we make. We therefore have the responsibility to guide, change and, when necessary, reject its applications. A reliable system that serves the human person is of greater value than one that is exceptionally powerful but whose consequences cannot be adequately foreseen or controlled.

Anton Tsitsulin avatar
Anton Tsitsulin@graph_·Oct 27

It's happened, I am a benchmark now

Today, we shared how Google’s Graph Mining and Graph Learning teams used Gemini 4 Argon with Teamwork agent in @GoogleAntigravity to solve a different class of internal engineering bottleneck: compressing months of low-level RDMA networking and custom TPU kernel development into hours. Specifically, improvements include accelerated graph algorithms with FarArray (up to 180x faster, 22x less RAM) and custom TPU kernels for Graph Neural Networks that run 8–15x faster than native JAX with massive HBM savings. For me, this update is a full-circle experience: Graph Mining contributed deeply to Gemini Data via deduping and data curation over time, and now Gemini is giving back by scaling and improving Graph Mining library! This is also a close collaboration amongst our Graph Mining and Teamwork teams, together with the Antigravity and Gemini teams. Congratulations! Read the full blog: antigravity.google/blog/gemi…

Petar Veličković avatar
⟳ Anton Tsitsulin retweeted
Petar Veličković@PetarV_93·Oct 26

We're also hiring Student Researchers in the Frontier Computation team @GoogleDeepMind, to work with @re_rayne & @IAmTimNguyen! Areas of interest: agentic optimisation, learning algorithms/simulations, OOD generalisation, AI4Physics, CFD, numerical math. Applying details below!

Anton Tsitsulin avatar
Anton Tsitsulin@graph_·Oct 26

Benchmarks, especially long horizon ones, are so harness dependent it’s crazy

Gemini 4 Argon is underrated 😱. When auditing the traces, we found that Pi Harness just ends the run whenever it hits a fallback (e.g. for security task) in 20% of the runs. We rerun Argon with Antigravity Harness, and now it beats Astra and sits right behind Opus 5.5. Congrats to the team @GoogleDeepMind and looking forward to the new ckpt.

0:05
Zachary Nado avatar
⟳ Anton Tsitsulin retweeted
Zachary Nado@zacharynado·Oct 22

The Gemini Large Scale Pretrain team is hiring a Research Engineer in London or MTV! The team works across the pretrain stack: research codebases, distributed training infra, data pipelines, tooling for training dynamics analysis, and most importantly using Gemini to accelerate our work! 🚀 If you're an engineer who has a strong "whatever it takes" attitude and is not afraid of spending tokens, apply here with some showcase work: google.com/about/careers/app… No particular qualifications needed, and we welcome early career applicants. Join if you want me to bug you with refactors to enable [new AGI compute multiplier]! 👀