Timur Garipov

AI research @ @MIT_CSAIL; ex-OpenAI

[Timur Garipov @tim_garipov]

We're hiring ML researchers and SWEs at Omnifold (https://t.co/zZqBi7JZSW) 🚀

Omnifold is building AI that transforms supply chain management.

Backed by Lightspeed & Kleiner Perkins ($28M raised), with customers from high-growth brands to $100B+ global logistics businesses.

🧵

[Philip Zmushko @fzmushko]

We’re excited to share our #ICML2026 paper on optimizer robustness in Async Pipeline Parallel LLM pretraining

Main message: with the right PP schedule, optimizer choice, and Error Feedback correction, Async PP can match sync training quality even at 10B MoE scale

[1/10]

[Timur Garipov @tim_garipov]

@vargastartup Congrats on the launch! 🚀

[Weight Space Symmetries @ ICML 2026 @weightsymmetry]

📢Excited to announce the Workshop on Weight-Space Symmetries @icmlconf! We welcome 4-page submissions analysing symmetries, their effects on training and model structure, and practical methods to utilize them.

Submission Deadline: April 24 (23:59 AoE)
#ICML2026

[Timur Garipov @tim_garipov]

All three roles are based in San Francisco (in-person).

If any of these sound like you or someone you know — reach out to Burke Schrauth at [email protected], DM me, or drop a comment below

[Timur Garipov @tim_garipov]

We're hiring ML researchers and SWEs at Omnifold (https://t.co/zZqBi7JZSW) 🚀

Omnifold is building AI that transforms supply chain management.

Backed by Lightspeed & Kleiner Perkins ($28M raised), with customers from high-growth brands to $100B+ global logistics businesses.

🧵

[Timur Garipov @tim_garipov]

⚙️ Infrastructure

Own the systems that make everything else possible — distributed systems, deployment, security, ML ops

Unique challenge: we train a custom model for every customer — infrastructure here is unlike anywhere else

Cloud/GPU, CI/CD, and AWS experience required

[Karina @karinanguyen]

Excited to release PostTrainBench v1.0!

This benchmark evaluates the ability of frontier AI agents to post-train language models in a simplified setting.

We believe this is a first step toward tracking progress in recursive self-improvement 🧵:

[Timur Garipov @tim_garipov]

I am at NeurIPS in San Diego this week. Let's meet up!

My brother Roman does impressive work in LLM inference and Reasoning. He is attending for the first time and presenting 2 papers at the conference! Come say hi at the posters (I will be there too).

[Roman Garipov @garipovroma]

I’m at NeurIPS in San Diego this week!
Presenting 2 papers from our Yandex Research LLM group — come say hi 👋
1️⃣ AutoJudge (poster) — lossy speculative decoding
🔗 https://t.co/uKiKQBLEgV
📚 https://t.co/oWUu89GNG8

2️⃣ Hogwild! Inference (spotlight poster) — concurrent collaborative reasoning with a single LLM instance
🔗 https://t.co/KxHUdtDnCe
📚 https://t.co/L227b3Kmtr

Excited to chat about LLMs, reasoning, and RL. Happy to talk research or grab a coffee — DM me!

[Pavel Izmailov @pavelizmailovai]

My Twitter account @Pavel_Izmailov has been compromised. I've lost access and am working with Twitter support to recover it. Please disregard any posts or DMs from that account.

[Jeremy Cohen @deepcohen]

Even with full-batch gradients, DL optimizers defy classical optimization theory, as they operate at the *edge of stability.*

With @alex_damian_, we introduce "central flows": a theoretical tool to analyze these dynamics that makes accurate quantitative predictions on real NNs.

[🎸 Rock History 🎸 @historyrock_]

Ozzy Osbourne has passed away at the age of 76. I still can't believe it.

A few weeks ago he was giving his final concert. I hope you came home, Ozzy ❤

[Alexander Theus @ ICML @TheusResearch]

1/ 🚨 New paper alert! 🚨
We explore a key question in deep learning:
Can independently trained Transformers be linearly connected in weight space — without a loss barrier?
Yes — if you uncover their rich symmetries.
📄 arXiv: https://t.co/wVoLYNzk0m

[Timur Garipov @tim_garipov]

@MillionInt https://t.co/UDRL4wXgaw

[Timur Garipov @tim_garipov]

Alex, congratulations to you and the team! Amazing work!

[Alexander Kirillov @_alex_kirillov_]

Finally, all the real-time conversation capabilities shown during the spring update are released! Video input was the last piece. I guess, time really flies with such an amazing team, it really feels like we demoed it just a couple of weeks ago. https://t.co/5gAoIBigM9