Helmholtz Institute for Human-Centered AI
@cpilab
We study human learning using tools from machine learning and improve machine learning using insights from cognitive science.
🚨 We're hiring! If you're excited about 🤖 ML/LLMs, 🧠 cognitive science, or 💭 computational psychiatry, come join us in Munich. Two fully funded PhDs @HelmholtzMunich: great mentorship, international vibe, lots of room to grow. 📅 May 16th 🔗 hcai-munich.com/PhDHCAI.pdf
🔬 Teaching AI to Understand Health: How #FoundationModels are Reshaping Biomedical Research At #HelmholtzMunich, researchers are developing and using large-scale AI to drive innovations in diagnostics, personalized medicine, and healthcare decision-making. 🧬 Prof. Fabian…
I recently defended my PhD thesis studying chunking in cognition and machine learning, working with Eric Schulz and Peter Dayan. @cpilab Here is a summary of my studies during the years in Tuebingen.
BREAKING: The first foundation model of cognition just dropped. It’s called Centaur, and it mirrors human behavior across 160+ psychology tasks. Here's everything you need to know:
We are seeking two independent Max Planck Research Groups working in scientific disciplines relevant to our institute. Applications should be sent in PDF format no later than 15th August 2025. Visit our call for applications on our institute website for detailed information!
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 human behavior. It predicts actual human behavior in held-out data & it generalizes to out-of-distribution tasks and experiments.
Excited to see our Centaur project out in @Nature. TL;DR: Centaur is a computational model that predicts and simulates human behavior for any experiment described in natural language.
Centaur – An AI That Thinks Like Us 🧠 Researchers at #HelmholtzMunich have developed an artificial intelligence model that simulates human decision-making with remarkable precision. 👉 t1p.de/4kjoy 🤖 The language model – called Centaur – was trained on over ten…
Excited to say our paper got accepted to ICML! We added new findings including this: models fine-tuned on a visual counterfactual reasoning task do not generalize to the underlying factual physical reasoning task, even with test images matched to the fine-tuning data set.
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:
A new study by @helmholtz_munich, @MPI_Kyb & @uni_tue shows GPT-4 excels in logic—but struggles with trust & teamwork. 🧠 Using a technique called Social Chain-of-Thought, researchers made it better at human-like interaction. #AI #GPT4 #SocialAI #GameTheory #TrustInAI
AI Meets Game Theory: New study reveals that while today’s AI is smart, it still has much to learn about social intelligence. Read more:🔗 t1p.de/ifuyh #AI #GPT4 #GameTheory #SocialAI
AI Meets Game Theory: New study reveals that while today’s AI is smart, it still has much to learn about social intelligence. Read more:🔗 t1p.de/ifuyh #AI #GPT4 #GameTheory #SocialAI
1/ 🚨 Updated preprint: “Generating Computational Cognitive Models using Large Language Models” 👥 Co-led by @milenamr7 with Marvin Mathony, Tobias Ludwig & @cpilab 📄 Check full paper here: arxiv.org/abs/2502.00879
Throughout the history of psychology, we have sought evidence of sophisticated cognitive capacities in non-human systems, including vertebrates 🐒 , insects 🐝 , cephalopods 🐙 , plants 🪴 , and bacteria 🦠
We're delighted to announce that Prof. Tali Sharot (@UCLPALS, @affectivebrain) has been awarded £3.5m as part of @wellcometrust Discovery Award scheme to investigate the 'joy of thinking’. buff.ly/V3oCDFR
Our paper (with @elifakata, @MatthiasBethge, @cpilab) on visual cognition in multimodal large language models is now out in @NatMachIntell. We find that VLMs fall short of human capabilities in intuitive physics, causal reasoning, and intuitive psychology. nature.com/articles/s4225…
Alignment is more than comparing similarity judgments! How well do pretrained neural networks align with humans in few-shot learning settings? Come check our poster #3904 at #NeurIPS on Wednesday to find out
Really amazing effort lead by @can_demircann! We measure how well neural networks align with how humans generalize about natural images in few-shot learning tasks. Vision-language models best capture how humans learn and generalize about natural images. Come chat at #NeurIPS!
Alignment is more than comparing similarity judgments! How well do pretrained neural networks align with humans in few-shot learning settings? Come check our poster #3904 at #NeurIPS on Wednesday to find out
Preprint alert! We explore 3 exploration tasks, testing if they measure a stable construct & its link to real-world exploration. Findings suggest improved robustness of latent factors compared to single-task estimates. Work with @mirkothm & @cpilab 🔗osf.io/preprints/psya… 🧵⬇️
In our latest preprint, we look at the reliability and validity of commonly used tasks to measurement human exploration behavior. Our results tell a cautionary tale of using such tasks to assess individual differences.
Preprint alert! We explore 3 exploration tasks, testing if they measure a stable construct & its link to real-world exploration. Findings suggest improved robustness of latent factors compared to single-task estimates. Work with @mirkothm & @cpilab 🔗osf.io/preprints/psya… 🧵⬇️
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!