Patrik Reizinger

@rpatrik96

🇭🇺 🇪🇺 Causality and Generalization in AI | @MPI_IS, @ELLISforEurope→@IVADO_Qc postdoc @Mila_Quebec | BiTS x @SPRIND @RenPhilanthropy | path2phd.substack.com

Tübingen, Germany

rpatrik96.github.io

Joined January 2018

349 Following

2,108 Followers

Replies Reposts Media

Pinned
Patrik Reizinger
@rpatrik96
Aug 3
The way we do science is changing. With AI-based tools, things we thought impossible became possible—and yet, in the non-representative sample I have seen, there is not enough experimentation among researchers. A thread ⬇️
Renaissance Philanthropy
@RenPhilanthropy
Jul 24
We’re delighted to introduce the @RenPhilanthropy and @SPRIND Big if True Science (BiTS) Accelerator 2026 cohort: eight exceptional scientists, technologists, and innovators developing ambitious R&D programs across AI for science, advanced materials, aging biology, nuclear
1
Patrik Reizinger
@rpatrik96
Sep 1
Clear writing has been one of my pet peeves, and I believe the emphasis is right. I also believe that we as a community should focus more on training scientists (incl. intelligent tool use) on this. Alas, I have built a tool with love and @claudeai ⬇️
Transactions on Machine Learning Research
@TmlrOrg
Aug 31
TMLR has refocused its acceptance criteria to place more emphasis on clear writing. Read more in this blog post: medium.com/@TmlrOrg/incre…
1
Patrik Reizinger
@rpatrik96
Aug 25
If you are also excited about principled mechanistic interpretability research, then we have good news! Our @NeurIPSConf in Sydney has amazing speakers, and an extended deadline of Sep 1st! Links below.
1
Patrik Reizinger
@rpatrik96
Aug 25
Oh, interesting! Our position paper at @icmlconf, led by @_shruti_joshi_ found similar things in mechanistic interpretability! Interactive demo: shrutij01.github.io/blog/causality… Paper: arxiv.org/abs/2602.16698
Kording Lab 🦖
@KordingLab
Aug 25
*The* problem in much of science including social neuro cognitive nutrition.
Quote
Misha Teplitskiy | Science of Science
@MishaTeplitskiy
Aug 24
The good news is that social scientists recognize it's important to make claims that are causal and not just associational. The bad news is that the evidence often does not support the claim.
1
Patrik Reizinger
@rpatrik96
Aug 19
When I started working on identifiable representation learning, I used the metrics everyone used. I assumed they quantify what they are "supposed to". I was wrong with that. Our @UncertaintyInAI poster (today), shows the limits of metrics. Led by @_shruti_joshi_ &an amazing team
Shruti Joshi
@_shruti_joshi_
Aug 18
How can we reliably measure whether a representation identifies latent factors? Can current identifiability metrics evaluate interpretability or unsupervised mechanism discovery? Our UAI 2026 paper finds systematic false positives and false negatives in these evaluations.
00:00