Filip Szatkowski

AI Scientist at and PhD Candidate at WUT

[leo 🐾 @synthwavedd]

Anthropic have officially invented benchminimizing

LMAO

[Piotr Miłoś @PiotrRMilos]

This week three new researchers joined the Mistral Warsaw office :). Let's achieve great things :).

Good luck @danie1marczak, @MozejkoMarcin, @f_szatkowski.

On the funny side, this means 75% monthly growth. Stay tuned for more good news soon.

[Liner @search_liner]

Liner is partnering with @spoticlr at #ICLR2026 — supporting Best Paper and Travel Awards for LLM research.

And to celebrate, we're giving away:
✈️ Round-trip flights + hotel to #ICML2026 in Seoul
🎁 $300 Liner Credits

Follow @search_liner + repost to enter by 4/27.

Liner is built for research workflows. Find papers, verify sources, and write with citations in one place.

See you in 🇧🇷 and 🇰🇷!

@iclr_conf @icmlconf

[Filip Szatkowski @f_szatkowski]

@AjPiorczynski @iclr_conf Thanks to coauthors Patryk Będkowski @devoto_alessio @jan_dubinski_ @PMinervini @s_scardapane Bartosz Wójcik

[Mikołaj Piórczyński @AjPiorczynski]

🇧🇷 Bom dia @iclr_conf ! This evening, together with @f_szatkowski, we’re presenting our paper “Universal Properties of Activation Sparsity in Modern Large Language Models”. Stop by Poster 912 in Pavilion 3 and let’s have a chat.

[Filip Szatkowski @f_szatkowski]

📷 https://t.co/DB1wMmUlK7

w/ Patryk Będkowski, @devoto_alessio, @jan_dubinski_, @PMinervini, @AjPiorczynski, @s_scardapane and Bartosz Wójcik

[Filip Szatkowski @f_szatkowski]

We are presenting "Universal Properties of Activation Sparsity in Modern Large Language Models" at ICLR 2026!

We ask a simple question: how sparse are modern LLMs, really — and does it matter? 👇

[Filip Szatkowski @f_szatkowski]

6/ Takeaway: activation sparsity isn't just a ReLU-era trick. It's a structural property of modern LLMs that scales favorably — and our work provides a principled way to harness it for efficiency and acceleration.

[Fundacja Quantum AI @qaifoundation]

🚀 Episode XXVIII of https://t.co/pPQHngM7RM is here! 🤖✨

📍 Where: Samsung R&D Institute Poland (Warsaw Spire, Plac Europejski 1)
🕕 When: 23 October, 18:00 (UTC+2)
🔗 Info: https://t.co/WEX3Rbj0Lm

📝 Register by: 22.10.2025 EoD → https://t.co/ovp38Fbo9R

[Wojtek Masarczyk @e7mul]

1/7 If Andrew Ng is right that the LR is the most important ML hyperparam, it's got some competition! We show that the softmax temperature is a game-changer in crafting NN representations. Often overlooked, it quietly governs generalization, collapse, and compression. A thread 👇

[Jan Dubiński @ ICML @jan_dubinski_]

🚨We’re thrilled to present our paper “CDI: Copyrighted Data Identification in #DiffusionModels” at #CVPR2025 in Nashville! 🎸❗️

"Was this diffusion model trained on my dataset?"
Learn how to find out:
📍 Poster #276
🗓️ Saturday, June 14
🕒 3:00 – 5:00 PM PDT
📜https://t.co/CLUrBtmQrI

TL;DR: We show that Membership Inference Attacks (MIAs) struggle to detect training data in SOTA Diffusion Models (DMs) and instead propose the first dataset inference method to achieve this goal.

#AI #MachineLearning #GenerativeAI #Copyright

[Bartłomiej Cupiał @CupiaBart]

First time being acknowledged in an OpenAI paper 👀

[Michał Bortkiewicz @ICML @m_bortkiewicz]

🚨Scaling RL
Most RL methods’ performance saturate at ~5 layers. In this work led by @kevin_wang3290, we crack the right configuration for scaling Contrastive RL and go beyond 1000 layers NNs! Deep NNs unlock emergent behaviors and other cool properties. Check out Kevin’s thread!

[ML in PL @MLinPL]

📢You’re invited to the Machine Learning Summer School on Drug and Materials Discovery (MLSS^D 2025)! 🚀🔬
🌟 MLSS^D is an intensive summer school focused on cutting-edge machine learning techniques and their applications in drug discovery and materials science.
By joining MLSS^D, you will:
✅ Learn from leading experts in ML and computational sciences
✅ Explore real-world case studies and practical applications
✅  Share your research with the world during a poster session
✅ Connect with researchers and professionals from around the world
🔗 Registration is now open! Register now: https://t.co/Epx27THRLt
🔥 When? July 1–6, 2025
📍 Where? Kraków, Poland
🌐 Find out more on our website: https://t.co/iPKCk04XOG
🚀 Don’t miss this unique opportunity to expand your knowledge and network in one of Europe’s most beautiful cities!
#MLSSD2025 #MachineLearning #DrugDiscovery #MaterialsScience #MLinPL

[Filip Szatkowski @f_szatkowski]

We are presenting "Exploiting Activation Sparsity with Dense to Dynamic-k Mixture-of-Experts Conversion" together with Bartosz Wójcik and @AjPiorczynski today at #NeurIPS2024 ! 🎉 If you are around and want to chat about efficiency, stop by poster 2304 at session 6 east at 4:30!