Eric Schulz

🤖 AI that thinks like us

Researchers at @www.helmholtz-munich.de have developed Centaur, an artificial intelligence model that can simulate human decision-making with remarkable accuracy.

@bethgelab.bsky.social @modirshanechi.bsky.social

Read the whole story 👉 bit.ly/4kCi280

Yes and yes.

An AI model (Llama 3.1 70B) fine-tuned on the results of 60,000 people in psychology experiments shows some real promise in using LLMs for studying and predicting human behavior.

It predicts actual human behavior in held-out data & it generalizes to out-of-distribution tasks and experiments.

What an exciting day for cognitive science with a double feature @nature.com including tiny models (www.nature.com/articles/s41...) and less tiny ones (www.nature.com/articles/s41...).

Thanks, Tobi.

What a day. Our foundation model of cognitive science is out in Nature:
www.nature.com/articles/s41...

@marcelbinz.bsky.social's thread:
bsky.app/profile/marc...

Centaur is featured by the NYT:
www.nytimes.com/2025/07/02/s...

We got featured in a Nature views:
www.nature.com/articles/d41...

Check out our new work on testing the limits of finetuning in improving vlms intuitive physics abilities.

Check out our new preprint in which we propose to use LLMs to generate new cognitive models that minimize scientific regret. This could be a viable approach to query foundation models to advance science more generally.

Generating Computational Cognitive Models using Large Language Models Computational cognitive models, which formalize theories of cognition, enable researchers to quantify cognitive processes and arbitrate between competing theories by fitting models to behavioral data....

Preprint update, co-led with @akjagadish.bsky.social, with @marvinmathony.bsky.social, Tobias Ludwig and @ericschulz.bsky.social!

What a sad day for academia all over the world. I was a postdoc at Harvard on a J1 visa. One of my students was going to start his postdoc there in a week. I can only imagine how awful the current situation and uncertainties around it must feel.

🔬 Foundation Models at #HelmholtzMunich drive advances in diagnostics, personalized medicine & cancer research.

👉 Read the full Highlight Story: t1p.de/d209f
✨ Discover more on our Foundation Models page: t1p.de/sx49u

#FoundationModels #AI

Our review with @ericschulz.bsky.social about generalization in humans and machines is out now in Current Directions in Psychological Science: doi.org/10.1177/0963...
Check it out!

One week left to apply, so please do... 😀

Our paper "Playing repeated games with large language models" just got published in Nature Human Behaviour! We explore how LLMs cooperate, coordinate, and adapt in social dilemmas. Check it out! Work led by @elifakata.bsky.social

email starting, "The U.S. National Science Foundation (NSF) has undertaken a review of its award portfolio. Each award was carefully and individually reviewed, and the agency has determined that termination of certain awards is necessary because they are not in alignment with current NSF priorites."

yesterday, my postdoc funding (salary and research funds) was cancelled by the National Science Foundation, effective immediately. I received the same generic, vaguely threatening, typo-ridden email as many of my colleagues who have had their awards terminated recently. (1/n)

#AIinScience: Ethical & Practical Challenges

🎤Interview with Dr. Marcel Binz, #HelmholtzMunich, on how Large Language Models are transforming scientific methods & the need of reshaping the scientific mindset:

👉 t1p.de/p82vj

@marcelbinz.bsky.social @ericschulz.bsky.social ‬@zeynepakata.bsky.social

At #ICLR2025, #HelmholtzMunich researchers are sharing advancements in #MachineLearning.

👉 t1p.de/8jfgn

@fabiantheis.bsky.social @carstenmarr.bsky.social @stes.io @zeynepakata.bsky.social @ericschulz.bsky.social @vincefort.bsky.social @iclr-conf.bsky.social

I had so much fun being a guest on the @behindthescreens.bsky.social podcast!
🤩

If you're curious about how we use games in research (and you understand German), check out their latest episode!

🚨 We're hiring! If you're excited about 🤖 ML/LLMs, 🧠 cognitive science, or 💭 computational psychiatry, come join us in Munich. Two fully funded PhDs @www.helmholtz-munich.de: tailored mentorship, international vibe, lots of room to grow.
📅 May 16th
🔗 hcai-munich.com/PhDHCAI.pdf

🧠 Keeping LLMs factually up to date is a common motivation for knowledge editing.

But what would it actually take to support this in practice at the scale and speed the real world demands?

We explore this question and really push the limits of lifelong knowledge editing in the wild.
👇

very happy to be presenting this at @cosynemeeting.bsky.social

Very grateful to be part of this.

Check out our review on generalization in humans and machines.

When we built the first cognitive foundation model, people said "it's only doing predictions". Here, we are working towards a foundation model of cognitive modeling. We want a tool that can make a useful guess as to what cognitive model, written in python code, could fit the inputted data.

Check out our new work on assessing the limits of finetuning in improving VLMs reasoning abilities.

BAMB! is happening again in 2025. Learn how to model your behavioural data, next to the beach in Barcelona! keynotes this year from @athenaakrami.bsky.social and Peter Dayan.

I've taken @arkrause.bsky.social's course on Probabilistic AI and @jonhue.bsky.social's script remains the most carefully written set of course notes I've encountered during my studies. Glad to see it released to the public! :)

Excited to see our review paper out in Annual Reviews. This paper nicely summarizes years of collaborative efforts working on Gaussian Process based theories of generalization, from function learning over developmental differences to compositionality.

What tech company will be the first to stand up and say that gutting the NSF is bad actually?

The economic argument seems pretty clear

Less NSF --> fewer PhD students --> fewer researchers --> smaller AI tech pipeline --> slower progress --> less competitive globally

If I study anxiety via threat-biased attention, the word "biased" gets me flagged. You can't design a study of humans without using at least one of the terms on the banned list, which means that biomedical, brain, social science research is now on ice in the USA