Sign in

micha heilbron

@mheilbron.bsky.social
1K followers 392 following 95 posts

Group leader at Max Planck Institute for Psycholinguistics @mpi-nl.bsky.social // Assistant Professor of Cognitive AI @UvA_Amsterdam. Cog-sci 🤝 AI 🤝 neurosci

PostsRepliesMedia
Reposted by micha heilbron
Heleen Slagter @haslagter.bsky.social · 02/10/2026
🚨 Job alert 🚨 Our department @vuamsterdam.bsky.social is hiring an Assistant Professor in Cognitive Neuroscience! Join our lovely community here in beautiful Amsterdam! 🧠🔥 Deadline: October 15 workingat.vu.nl/vacancies/as...
workingat.vu.nl
Vacancy — Assistant Professor in Cognitive Neuroscience
Join the Department of Experimental and Applied Psychology (Vrije Universiteit Amsterdam) as Assistant Professor in cognitive neuroscience.
15631
Reposted by micha heilbron
Enny van Beest @ennyvb.bsky.social · 04/09/2026
Textbook: tracking position during navigation is primarily the job of the hippocampus. Recent findings suggest a more complicated story. Recording >20k neurons brainwide, we indeed find spatial representations in every region! With @kenneth-harris.bsky.social and @carandinilab.net
Brainwide representations of a virtual corridor
416852
micha heilbron @mheilbron.bsky.social · 20/08/2026
Tracking entities is a key building block of understanding When measured as story comprehension rather than puzzle-solving, this ability emerges in models far smaller than previously thought -- and at scale, models far exceed human performance 9/9, fin arxiv.org/abs/2608.18083
arxiv.org
Entity tracking emerges in sub-billion parameter language models and exceeds human performance in naturalistic narratives
Understanding language requires tracking entities across discourse - i.e., knowing where things are and how they change, even when not explicitly stated. Whether language models perform such tracking ...
031
micha heilbron @mheilbron.bsky.social · 20/08/2026
Is this actual tracking, or just word associations (e.g. that keys are often in boxes)? We re-ran it with pseudowords and bizarre objects (kidney, silence) and find that the accuracy held -- implying that models rely on structured tracking, not lexical associations 8/9
110
micha heilbron @mheilbron.bsky.social · 20/08/2026
Where does this ability come from? We compared OLMo base and instruct models, and see that instruction tuning improves explicit question answering, but not the implicit measure, and thus not entity tracking itself 7/9
100
micha heilbron @mheilbron.bsky.social · 20/08/2026
In humans, tracking is robust but imperfect, and accuracy falls with the complexity of the scene, not its length or entity recency. In models, tracking emerges with scale and on the implicit measure, robust tracking occurs already in very small models of 410M parameters! 6/9
100
micha heilbron @mheilbron.bsky.social · 20/08/2026
To address this, we designed a new, controlled task measuring on-the-fly entity tracking in readable narratives of varying complexity We tested both humans and open model families (Pythia, OLMo 2), with implicit (probability read-out) and explicit (generation) evaluation 5/9
100
micha heilbron @mheilbron.bsky.social · 20/08/2026
However, their task was abstract and complex — more like solving a puzzle than following a story — and didn’t compare against humans So maybe the high bar reflects the reasoning nature of that task -- not entity tracking itself, as used in natural language? 4/9
100
micha heilbron @mheilbron.bsky.social · 20/08/2026
Prior work (by @sebschu.bsky.social, @najoung.bsky.social ) found this ability only in large, code-trained models: for instance, not even text-only GPT-3 (175B), but the later, code-specialised GPT-3.5 So: only models that code well can robustly follow language?
100
micha heilbron @mheilbron.bsky.social · 20/08/2026
Understanding language requires building an internal model of the situation a text describes This allows comprehenders to track entities – to know where they are and how they change, even when not explicitly stated Do language models do this in a human-like manner? 2/9
110
micha heilbron @mheilbron.bsky.social · 20/08/2026
Paper: arxiv.org/abs/2608.18083 Thread below 👇
arxiv.org
Entity tracking emerges in sub-billion parameter language models and exceeds human performance in naturalistic narratives
Understanding language requires tracking entities across discourse - i.e., knowing where things are and how they change, even when not explicitly stated. Whether language models perform such tracking ...
110
micha heilbron @mheilbron.bsky.social · 20/08/2026
New preprint, w/ Karolina Drożdż Understanding language requires internally modelling what's happening where, even when unstated We built a new task to measure this ability in language models and people, and find it emerges at model scales far smaller than previously thought
1207
micha heilbron @mheilbron.bsky.social · 19/08/2026
amazing Ellie! many congratulations!
010
Reposted by micha heilbron
Sam Gershman @gershbrain.bsky.social · 18/08/2026
@arthurpr4t.bsky.social has a new preprint with important results on a famous psychophysical law (Weber's law). It isn't, in fact, a law, because it can be broken. A more fundamental principle (efficient coding) shows when and why Weber's law holds true. www.biorxiv.org/content/10.6...
biorxiv.org
Efficient coding makes and breaks Weber's law
Weber's law is a rare quantitative regularity in psychology, yet its origins remain debated. Here we provide causal evidence that it arises from the more fundamental principle of efficient coding. Thi...
413041
micha heilbron @mheilbron.bsky.social · 15/08/2026
hmm i dunno what service that is, but the only serious and actually working detector -- pangram -- marks it as human
4160
Reposted by micha heilbron
William Timkey @wtimkey.bsky.social · 10/08/2026
Why do we breeze through some sentences, but others make us slow down and reread? A popular answer is predictability: unexpected words are harder to process. In our new @pnas.org article, we used LMs as models of human prediction to ask how far this explanation can actually go🧵
1276
micha heilbron @mheilbron.bsky.social · 20/07/2026
Full, definitive paper here: direct.mit.edu/tacl/article...
direct.mit.edu
Human-like Fleeting Memory Improves Language Learning but Impairs Reading Time Prediction in Transformer Language Models
Abstract. Human memory is fleeting. As words are processed, the exact wordforms that make up incoming sentences are rapidly lost. Cognitive scientists have long believed that this limitation of workin...
010
micha heilbron @mheilbron.bsky.social · 20/07/2026
Belated, but still happy to see our paper (with @drhanjones.bsky.social) on fleeting memory transformers is out in TACL! We find that giving language models human-like memory decay *improves* language learning, while, unexpectedly, impairing human reading time prediction Follow up results soon!
191
Reposted by micha heilbron
Nicole Rust @nicolecrust.bsky.social · 18/07/2026
Wonderful to see this! For (controversial?) context. There’s long been an argument that what brains & ANNs are doing cannot be fathomed beyond the meta like (eg) architecture, learning rules and such. And there’s a counter-idea: ”let’s try?”. When we stumbled on the correlates of memorability /1
2296
micha heilbron @mheilbron.bsky.social · 18/07/2026
Hadn’t seen this yet — very interesting! @davogelsang.bsky.social
010
micha heilbron @mheilbron.bsky.social · 16/07/2026
www.sciencedirect.com/science/arti... (Open access link: authors.elsevier.com/c/1nRZk,H2pb...)
sciencedirect.com
Representational magnitude as a geometric signature of image and word memorability
What makes some stimuli more memorable than others? Recent work has shown that image memorability is predicted by the magnitude of population response…
180
micha heilbron @mheilbron.bsky.social · 16/07/2026
What makes some stimuli more memorable than others? In a new paper w/ @davogelsang.bsky.social, we show that the magnitude of a stimulus's ANN representation predicts both image and word memorability Stimuli that activate more features, more strongly, leave a stronger memory trace Out now in JML⬇️
34411
Reposted by micha heilbron
Arvind Narayanan @randomwalker.bsky.social · 15/07/2026
I had the honor of giving a keynote at the International Conference on Machine Learning last week. I addressed the widespread anxiety about how we should adapt as AI capabilities increase. I was thrilled by the talk’s reception, so I have made my slides available www.cs.princeton.edu/~arvindn/tal...
cs.princeton.edu
What will be left for us to work on?
ICML 2026 invited keynote — slides and edited transcript, presented click-by-click as delivered. Arvind Narayanan, Princeton University.
614048
Reposted by micha heilbron
Max Planck Institute for Psycholinguistics @mpi-nl.bsky.social · 23/06/2026
Human-like fleeting memory improves language learning but impairs reading time prediction in transformer language models. New paper by Abishek Thamma & @mheilbron.bsky.social doi.org/10.1162/TACL.a.688
082
Reposted by micha heilbron
de Volkskrant @volkskrant.nl · 04/06/2026
Als geen ander wist Lieke Marsman (1990-2026) het allerzwaarste licht te maken
volkskrant.nl
Als geen ander wist Lieke Marsman (1990-2026) het allerzwaarste licht te maken
36112
Reposted by micha heilbron
Sebastiaan Mathôt @cogsci.nl · 22/05/2026
Nu online te bekijken! Een Wereld vol Denkers 🌍🌿🤖🧠🐝 bij Studium Generale #Maastricht. Dit was echt een hele leuke avond! Met @mheilbron.bsky.social! @maastrichtu.bsky.social @uitgeverijbalans.bsky.social #wetenschap #boeken #biologie #psychologie #AI www.youtube.com/watch?v=QKjh...
youtube.com
Lezing | Een wereld vol denkers: mens, dier, plant en AI | Sebastiaan Mathôt & Micha Heilbron
YouTube video by Studium Generale Maastricht University
142
Reposted by micha heilbron
ines-schoenmann.bsky.social @ines-schoenmann.bsky.social · 27/04/2026
New peer-reviewed paper w/ @mheilbron.bsky.social, @predictivebrain.bsky.social & Jakub Szewczyk! Pre-onset brain encoding has been taken as evidence that brains–like LLMs–predict upcoming words. We show that the same signatures arise in systems that cannot predict. (elifesciences.org) (1/8)
312050
Reposted by micha heilbron
Sebastiaan Mathôt @cogsci.nl · 22/04/2026
Ik heb er zin in! Morgen zijn @mheilbron.bsky.social en ik bij @maastrichtu.bsky.social voor een avond vol #wetenschap, #biologie, #psychologie en #AI! 🧠🐝🌿🤖 Meld je aan via www.maastrichtuniversity.nl/nl/events/ee... #maastricht
maastrichtuniversity.nl
Een wereld vol denkers: mens, dier, plant en AI - Agenda - Maastricht University
021
micha heilbron @mheilbron.bsky.social · 15/04/2026
Nijmegen friends: Tomorrow (10–12) I'll be debating Pim Haselager at a Donders Session on the thesis: "Artificial neural network models are adequate mechanistic models of the mind" I'm defending, he's opposing. Should be fun. Come join us! www.ru.nl/en/donders-i... @dondersinst.bsky.social
ru.nl
Donders Session - 16 April | Radboud University
Donders Debate with Micha Heilbron and Pim Haselager: Artificial neural network models are adequate mechanistic models of the mind
1140
Reposted by micha heilbron
Sebastiaan Mathôt @cogsci.nl · 13/04/2026
23 april geven @mheilbron.bsky.social en ik een lezing in het mooie #Maastricht over Een wereld vol denkers. Een avond vol verhalen over het denken en doen van mens, dier, plant en AI! 🧠🐝🌿🤖 Ik hoop jullie daar te zien! www.maastrichtuniversity.nl/nl/events/ee... #wetenschap #psychologie #biologie
maastrichtuniversity.nl
Een wereld vol denkers: mens, dier, plant en AI - Agenda - Maastricht University
052
Reposted by micha heilbron
Limor Raviv 🐘🤗🦒🍄🦄 @limorraviv.bsky.social · 08/04/2026
I'm hiring! 📢 Fully funded 4-year PhD position in Language Evolution using Communication Games at @mpi-nl.bsky.social. Come work with me on how different social pressures shape the evolution of new communication systems in the lab! Deadline for application is May 18th! share.google/fGTKbFS4v4Gb...
share.google
Fully funded 4-year PhD position in Language Evolution using Communication Games | Max Planck Institute
03327
micha heilbron @mheilbron.bsky.social · 08/04/2026
Well not necessarily without prediction, but recent evidence pointed that predictabilty effects were mostly high-level journals.plos.org/ploscompbiol...; direct.mit.edu/imag/article...) but our new work shows an interesting twist, it seems to depend on eccentricity (or sensory reliability)
journals.plos.org
000
micha heilbron @mheilbron.bsky.social · 08/04/2026
Classic predictive coding: V1 predicts low-level features, higher areas high-level. But recent studies + AI models suggest prediction happens at higher levels of abstraction. Who's right? In new work w/ @wiegerscheurer.bsky.social we find that both are – distinct regimes across the visual field
1256
Reposted by micha heilbron
Wieger Scheurer @wiegerscheurer.bsky.social · 04/04/2026
New preprint! w/ @mheilbron.bsky.social We found that, even during simple natural scene viewing, human visual cortex predicts—hierarchically in central vision and at higher levels peripherally—reconciling classical predictive coding with recent evidence from animal models and AI (e.g. JEPA) (1/10)
13311
Reposted by micha heilbron
Athena Akrami @athenaakrami.bsky.social · 07/04/2026
Academic friends, It's beyond heartbreaking to watch what's unfolding in Iran & the region. A few of us drafted an open letter calling for protection of civilians & of educational, research, medical & cultural institutions. Please read & sign if you agree: sites.google.com/view/protect... #IranWar
sites.google.com
Protect Academic Life in Iran
We, the undersigned academics and researchers from around the world, express our profound concern over recent military strikes on Iran, the retaliatory responses, and the reported impact on civilian l...
28450
micha heilbron @mheilbron.bsky.social · 30/03/2026
Interested in pursuing a PhD in NLP/cog-sci? Studying language learning in LMs from the perspective of human language acquisition? Few more days to apply!!
071
Reposted by micha heilbron
Stéphane d’Ascoli @sdascoli.bsky.social · 26/03/2026
🚨 We're very happy to introduce TRIBE v2: a foundation model of the brain's responses to sight, sound & language. 📄 Paper: ai.meta.com/research/pub... ▶️ Demo: aidemos.atmeta.com/tribev2/ 💻 Code: github.com/facebookrese... 🤗 Model: huggingface.co/facebook/tri...
36320
Reposted by micha heilbron
micha heilbron @mheilbron.bsky.social · 10/03/2026
📢 PhD position in Developmental Language Modelling (PLZ RT) What can human language acquisition teach us about training language models? Join us as a PhD! mpi.nl/career-education/vacancies/vacancy/fully-funded-4-year-phd-position-developmental-language @carorowland.bsky.social @mpi-nl.bsky.social
12935
micha heilbron @mheilbron.bsky.social · 10/03/2026
mpi.nl/career-education/vacancies/vacancy/fully-funded-4-year-phd-position-developmental-language
mpi.nl
Fully Funded 4-Year PhD Position In Developmental Language Modelling | Max Planck Institute
010
micha heilbron @mheilbron.bsky.social · 10/03/2026
📢 PhD position in Developmental Language Modelling (PLZ RT) What can human language acquisition teach us about training language models? Join us as a PhD! mpi.nl/career-education/vacancies/vacancy/fully-funded-4-year-phd-position-developmental-language @carorowland.bsky.social @mpi-nl.bsky.social
12935
micha heilbron @mheilbron.bsky.social · 05/03/2026
📢 PhD position in the NeuroAI of Language Why can LLMs predict brain activity so well? We're hiring a PhD student to find out -- AI interpretability meets neuroimaging Deadline March 20 Please RT 🙏 👇 mpi.nl/career-education/vacancies/vacancy/fully-funded-4-year-phd-position-neuroai-language
25239
micha heilbron @mheilbron.bsky.social · 27/02/2026
yes i will be around -- let's do it!
110
micha heilbron @mheilbron.bsky.social · 27/02/2026
(I'll keep a part-time affiliation with the @uva.nl as Assistant Professor of Cognitive AI, continuing to teach all things AI and the brain/mind, so I'll still be around in Amsterdam)
060
micha heilbron @mheilbron.bsky.social · 27/02/2026
Job update: Next week I start as a group leader at the Planck Institute for Psycholinguistics in Nijmegen @mpi-nl.bsky.social 🧠 Building the Language and Predictive Computation group -- using LLMs to model language in the mind/brain, and vice versa. Hiring soon!
3552
Reposted by micha heilbron
Andrew Lampinen @lampinen.bsky.social · 18/02/2026
What is the relationship between memorization and generalization in AI? Is there a fundamental tradeoff? In infinitefaculty.substack.com/p/memorizati... I’ve reviewed some of the evolving perspectives on memorization & generalization in machine learning, from classic perspectives through LLMs.
infinitefaculty.substack.com
Memorization vs. generalization in deep learning: implicit biases, benign overfitting, and more
Or: how I learned to stop worrying and love the memorization
415629
micha heilbron @mheilbron.bsky.social · 03/02/2026
Interesting convergence: The trick that made predictive self-supervised vision models work seems to be what the brain was doing all along w/ @predictivebrain.bsky.social: visual cortex is most sensitive to high-level prediction errors -- even in V1 Now published: journals.plos.org/ploscompbiol...
journals.plos.org
Higher-level spatial prediction in natural vision across mouse visual cortex
Author summary How does the brain make sense of the constant stream of visual information? A popular theory suggests the brain is not a passive receiver but an active predictor, constantly generating ...
0245
Reposted by micha heilbron
Alexander Huth @alexanderhuth.bsky.social · 05/01/2026
This paper had a pretty shocking headline result (40% of voxels!), so I dug into it, and I think it is wrong. Essentially: they compare two noisy measures and find that about 40% of voxels have different sign between the two. I think this is just noise!
8241100
micha heilbron @mheilbron.bsky.social · 19/11/2025
so nice to see this out sush!!
110
Reposted by micha heilbron
Sushrut Thorat @martisamuser.bsky.social · 18/11/2025
🚨New Preprint! How can we model natural scene representations in visual cortex? A solution is in active vision: predict the features of the next glimpse! arxiv.org/abs/2511.12715 + @adriendoerig.bsky.social , @alexanderkroner.bsky.social , @carmenamme.bsky.social , @timkietzmann.bsky.social 🧵 1/14
arxiv.org
Predicting upcoming visual features during eye movements yields scene representations aligned with human visual cortex
Scenes are complex, yet structured collections of parts, including objects and surfaces, that exhibit spatial and semantic relations to one another. An effective visual system therefore needs unified ...
38528
micha heilbron @mheilbron.bsky.social · 07/11/2025
archive.ph/smEj0 (or, unpaywalled 🤫)
archive.ph
020