🔥Telegram İfşa

Come work with Can Demircan and me on semantic label smoothing (it's cool, I promise)!

Object slots are great for compositional generalization, but can models without these inductive biases learn compositional representations without supervision too? Yes! Unsupervised learning on object videos yields entangled, yet compositional latent codes for objects!

Excited to announce Centaur -- the first foundation model of human cognition. Centaur can predict and simulate human behavior in any experiment expressible in natural language. You can readily download the model from Hugging Face and test it yourself: huggingface.co/marcelbinz/Lla…

🚨Preprint alert🚨 In an amazing collaboration with Gruaz Lucas, Sophia Becker, & J Brea, we explored a major puzzle in neuroscience & psychology: *What are the merits of curiosity⁉️* osf.io/preprints/psya… 1/7

Super excited to be going to #NeurIPS to present new work on softly state-invariant world models! We introduce an info bottleneck making world models represent action effects more consistently in latent space, improving prediction and planning! Reach out if you want to meet!

Super happy that this has been accepted to ICLR 2026 ! Me and Can Demircan will be there to talk about what we can learn about in-context learning using SAEs See you in Singapore 🇸🇬

Happy to say we made the cover too! (Elif Akata, Helmholtz Institute for Human-Centered AI)

In previous work we found that VLMs fall short of human visual cognition. To make them better, we fine-tuned them on visual cognition tasks. We find that while this improves performance on the fine-tuning task, it does not lead to models that generalize to other related tasks:

New in PNASNews: doi.org/10.1073/pnas.2… We study how humans explore a 61-state environment with a stochastic region that mimics a “noisy-TV.” Results: Participants keep exploring the stochastic part even when it’s unhelpful, and novelty-seeking best explains this behavior.

Turns out some induction heads are very good at learning hierarchical sequences. Check out our work below on the mechanisms behind this!

🚀 We are hiring! 🚀 🔍 Join us as a Postdoctoral Researcher (fully-funded) at the Helmholtz Institute for Human-Centered AI in Munich. Helmholtz Munich | @HelmholtzMunich Helmholtz AI ELLIS