Timo Kaufmann

Joint work with Yannick Metz, Daniel Keim, and Eyke Hüllermeier.

Benefits: improved reward model generalization, better data efficiency, and stronger policies. Looking forward to seeing you at the poster!

Paper and more: timokaufmann.com/responserank/

The key insight is that these signals only need to be locally valid and relative (e.g., within one annotator's comparisons). No need to model the exact relationship to strength. Just rank which comparisons are stronger.

The core idea: Not all preferences are equal. ResponseRank learns preference strength from implicit signals in your data, like inter-annotator agreement, stated confidence, or response times.

Presenting ResponseRank at #NeurIPS2025! Come by poster #405 at 4:30pm today if you're in San Diego 👋

Just noticed the key deadlines for #ICLR2026 out! PSA for everyone else who's been waiting.

Full paper: Sept 24 AoE.

A bit late to post on BlueSky, but I had a great time at our poster session. Very cool to see so much interest in ICAI :)

🕵🏻💬 Introducing Feedback Forensics: a new tool to investigate pairwise preference data.

Feedback data is notoriously difficult to interpret and has many known issues – our app aims to help!

Try it at app.feedbackforensics.com

Three example use-cases 👇🧵

Currently visiting @arduin.io in Cambridge. I didn't realize it's this beautiful!

Do I know anyone here that I haven't met up with yet?

👋 I'll start cross-posting from twitter for now.