ICLR2026 has ended, and it was my very first scientific conference where I was presenting a paper!
A few notes from Rio: what I learned, what I’m bringing back to my research, and some thoughts on the way.
Presenting the Poster#
I presented MASS, my very first paper on MoErging, at the main poster session. I really liked the format, and the role a well-designed, attrctive, poster plays. Crafting one turned out to be a creative side of the work I hadn’t considered before.
A few things stood out:
- I made a rookie mistake on my very first pitch of the poster: I jumped in assuming the person already knew what model merging was. They didn’t.
- Some people who had already read the paper stopped by just to say congrats, and that meant a lot.
- Others didn’t just stop by, they wanted to understand the work deeply, and they asked very sharp, very specific questions.
The Workshop Re³-Align#
The Re³-Align workshop, beyond the interesting talks, brought a noticeably different distribution of attendees to the poster sessions. With a smaller crowd, questions got more profound: there was actually time to stop, think, and come back with new insights.
What I’m Taking Home#
- Research direction: for sure I would like to dive more into the LLM formal proofing world.
- Community: I met tens of intrsting people, I hope to see at next events.
- Experience: I’ve learnt a lot about presenting a poster effectively!
Resources#
- MASS paper: see the publications page.
- Re³-Align workshop: representational-alignment.github.io/2026.
- ICLR 2026: iclr.cc/Conferences/2026.
