Ilya Lasy
Currently doing PhD at TU Wien, Austria. Right now I’m completely fascinated by mechanistic interpretability of LLMs, so my research interest lies in that domain.
I am not particularly fond of LLMs, but I feel like we’ve achieved something quite incredible with quite a simple technology. Don’t get me wrong, I don’t want to belittle the achievements of all these remarkable researchers, but I believe that just scaling the next token prediction is not going to lead us to the brightest future.
Anyways, I love digging around in wires & circuits of modern LLMs with a hope to understand their “emergent capabilities” and contribute to better architectures in the future.
Besides all that, I’m a huge nerd, storytelling enjoyer and (indie-)gamer. TTRPGs, LARPs, TCGs (mostly Magic) - you name it. I’ve been DMing for quite a while now - so if interested in a game - just let me know.
Feel free to reach out, would be happpy to chat about mech interp, art, music, games, singularity, anything else really too.
news
| May 02, 2026 | Our paper on interpreting expert routing in MoEs got accepted at ICML 2026! |
|---|---|
| Jun 11, 2025 | My paper got accepted at The First Workshop on Large Language Model Memorization at ACL 2025! |
| Jan 24, 2024 | Co-organised a local hackathon to detect LLMs’ hallucinations. Published a shared-task paper after that, was fun! |