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.
Top: Modality shift in block instruction: R1: Take the green block and put it on the left side of the grid. A hand is holding an imaginary piece toward the left column of a 2×2 grid; label reads Redundant position and orientation. R4: the green block pointing this way. A hand is pointing near the bottom left cell with an arrow showing movement toward the top left cell; label reads Complementary position and orientation. Target tower: a 3 block green and red C-shape tower on a 2x2 grid. Bottom: Modality shift in tower instruction: R1: they are going to form a C-shape. A c-shape hand pose with the index and thumb is shown far from the grid; the label reads No information about position or orientation. R4: Put the C on the left side, facing away from you. Right hand shows the C shape facing away, and left hand with the palm open indicates placement on the left side; labels read Redundant position and orientation.
People form ad hoc conventions, establishing linguistic & gestural abstractions, and shift information across speech and gesture to communicate more efficiently over time.
We study this in our #CHI2026 paper, led by Kiyosu Maeda with @judithfan.bsky.social @rdhawkins.bsky.social and team
🧵👇
1/4
Final paper of my PhD 🤗
www.nature.com/articles/s44...
There is growing interest in how cognitive control may improve value-based decision making.
However, we find that a recent paper overestimated the role of control in their task, leading to erroneous interpretations of dACC recordings.
How do diverse context structures reshape representations in LLMs?
In our new work, we explore this via representational straightening. We found LLMs are like a Swiss Army knife: they select different computational mechanisms reflected in different representational structures. 1/
1/7 Can infants recognise the world around them? 👶🧠 As part of the FOUNDCOG project, we scanned 134 awake infants using fMRI. Published today in Nature Neuroscience, our research reveals 2-month-old infants already possess complex visual representations in VVC that align with DNNs.
Our paper (w/ @bodowinter.bsky.social and @mperlman.bsky.social) is finally out, officially 🥳. In it, we set ourselves the lofty goal of defining iconicity, focusing on its subjectivity, context-dependence, and gradability. Let us know if you agree with our definition? 🤔
doi.org/10.1093/oxfo...
🎭 How do LLMs (mis)represent culture?
🧮 How often?
🧠 Misrepresentations = missing knowledge? spoiler: NO!
At #CHI2026 we are bringing ✨TALES✨ a participatory evaluation of cultural (mis)reps & knowledge in multilingual LLM-stories for India
📜 arxiv.org/abs/2511.21322
1/10
With some trepidation, I'm putting this out into the world:
gershmanlab.com/textbook.html
It's a textbook called Computational Foundations of Cognitive Neuroscience, which I wrote for my class.
My hope is that this will be a living document, continuously improved as I get feedback.
🚨 New preprint 🚨
Excited to share new work led by @tikhomirova.bsky.social, presenting a large-scale evaluation of the accessibility of psycholinguistic information in transformer language models.
Preprint: arxiv.org/abs/2601.03798
Elmar Unnsteinsson (@eunnsteins.bsky.social) and I have a new paper forthcoming in Noûs.
"Genre and Conversation"
We show how to generalize the classic pragmatic theories to conversational genres that aren't factual, cooperative, committal information exchanges.
philpapers.org/rec/ELMGAC-2
New paper from the IMC lab! I am very excited about this one. For years, I have been arguing that one of the main claims of the so-called "simulation heuristic" is likely not true for episodic counterfactual thinking, namely that the harder it is to mentally simulate it, the less plausible (1/n)
If you've missed this piece about the different modes of empiricism in computer science versus the social sciences, I can highly recommend it. doomscrollingbabel.manoel.xyz/p/the-empiri...
New paper w/ @ryskin.bsky.social in Open Mind!
Words change: “broadcast” once meant scattering seeds; “tweet” was just a bird sound. Do older adults keep earlier meanings, or update as language evolves?
Our new paper investigates how semantic representations differ across age groups. 🧵👇
The question is an open one and it depends on us. The would-be destroyers of science #Trump #Vought #RFKJr #Bhattacharya #Memoli are just men. Zealots, fools, charlatans, opportunists. There are more of us than there are of them. www.nature.com/articles/d41...
New study out in Neuron: doi.org/10.1016/j.ne.... This work led by Zaid Zada uses fMRI hyperscanning of real dyads to show that speaking and listening rely on shared neural systems; and that conversation recruits unique brain processes that aren't observed in passive comprehension.
Great to see this paper on sensory expectations in motor control from @jonathanamichaels.bsky.social and @andpru.bsky.social out in Nature today!
www.nature.com/articles/s41...
#neuroscience 🧪
Thom Scott-Phillips presents a novel analysis of people's spontaneous intuitions about sentence acceptability "grounded in theoretical and empirical knowledge from cognitive linguistics, cognitive psychology and evolutionary approaches to the mind." cadernos.abralin.org/index.php/ca...
Now out in TopiCS as "Simulating Symbolic Evolution in the Lab: Potentials and Implications of Using Transmission Chains to Study Early Symbolic Behavior at the Emergence of Homo sapiens" doi.org/10.1111/tops... Thread below! w @felixthehauskat.bsky.social & many others.
We're hiring at the Associate Professor level in Social Psychology! Join us in our vibrant and wonderful department and area. Contact Markus Brauer, search committee chair, for details (markus.brauer@wisc.edu). Job ad here: shorturl.at/nnvv8
Bichan Wu (@bichanw.bsky.social) & I wrote a tutorial paper on Reduced Rank Regression (RRR) — the statistical method underlying "communication subspaces" from Semedo et al 2019 — aimed at neuroscientists.
arxiv.org/abs/2512.12467
Now in press at Topics in Cognitive Science! We review children's understanding of different kinds of visual media across contexts, and argue that this work has important implications for childhood learning and assessment tools @cogscisociety.bsky.social
onlinelibrary.wiley.com/doi/10.1111/...
www.science.org/doi/10.1126/...
damn this is so very clever!
"... we investigated how the brain categorizes stimuli that are not linearly separable in the physical world ... The sensory manifold was ... expanded into a seven-dimensional perceptual manifold..."
The Press and @openmindjournal.bsky.social are pleased to announce a partnership with Lyrasis through the Open Access Community Investment Program (OACIP).
Learn how your institution can support this initiative to continue providing the latest #cogsci research—free of charge—here: bit.ly/452nMma
New preprint in advance of a Phil Trans paper. Outlining a theoretical argument bridging Bayesian causal learning and empowerment in reinforcement learning. And empirical data that kids do too!
arxiv.org/abs/2512.08230
Sign on bathroom wall: (In red) IF TOILET IS STUCK, DO NOT FLUSH AGAIN OR TOILET WILL OVERFLOW. (Picture of overflowing toilet) (In green) GET A STAFF: TO HELP OR USE THE PLUNGER. (Two pictures of plungers)
This is a nice in-the-wild example of a rather complex dynamic semantic speech act: a conditional with a factual antecedent, an imperative consequent which itself embeds a factual consequent, producing a warning, and is itself conjoined to another disjunctive imperative consequent.
🐦🐦 #philsky
New paper on lexical variation in Swedish Sign Language with Swedish colleagues.
Using a combo of elicitation (in-person), survey (online) & corpus data, we look at some changes in lexical choices over time & discuss methods for measuring variation of variation
#linguistics
doi.org/10.16995/glo...
A case in point: Dave Freedman has found abstract feature-based categories in LIP, supposedly a "where" not a "what" area. He also found them in the superior colliculus, which is supposed to be an oculomotor structure and have nothing to do with visual categories!
www.nature.com/articles/s41...
If you speak whale, you may know that sperm whales communicate with click vocalizations that they group into units called codas. A new study in @openmindjournal.bsky.social shows that within these codas appear to be vowels & diphthongs, used similarly to humans: direct.mit.edu/opmi/article...
