***4-year Career Development Research Fellowship in Psychology now available at Oxford!***
Open to all areas of Psychology, including Cognitive/Behavioural Neuroscience.
PhD in last three years, or be about to submit (some exceptions apply).
Deadline: 21st May.
www.sjc.ox.ac.uk/discover/vac...
New paper 🌟out now in Nat. Neurosci.: www.nature.com/articles/s41.... With advisors @davidpoeppel.bsky.social and @Nai!
We show that, while LLMs are optimized to predict the next word, the human brain modulates prediction efficiency by strategically grouping words into constituents. 1/n
More 👇
An image by Sketchplanations, illustrating Geroge Box's quote, "All model are wrong, but some of useful"
New post! Are you following @sketchplanations.bsky.social ? He is great at explaining complex ideas via text and images, and some of those complex ideas are statistical.
Good news everyone 🥳 Our (w @vincentab.bsky.social) primer on models as prediction machines (with the marginaleffects package) is finally officially published!>
journals.sagepub.com/doi/10.1177/...
🎯New preprint!
The resource-rational dynamics of evidence accumulation
www.biorxiv.org/content/10.6...
Evidence accumulation isn’t stationary.
It is dynamically controlled to trade off information gain against cognitive effort.
With Jiang Mao, Tobias Donner @donnerlab.bsky.social, Alan Stocker
Thanks for pointing out the paper. It's unfortunate that we didn't come across this work in our review of the literature. I am constantly reminded of the depth and breadth of work on cognitive maps.
Effect sizes are increasingly promoted in psychological science, sometimes even considered the primary outputs of quantitative research. While effect sizes are essential for statistical purposes related to open science efforts (replication planning, meta-analysis), their widespread interpretation as markers of practical relevance is problematic. Specifically, this article argues that effect sizes approximate practical relevance in applied research, which aims to mirror real-world environments, but not in theory-testing research, which aims to isolate causal mechanisms. Three reasons for this limitation are outlined. First, theory-testing effects often differ fundamentally from practical effects the theory aims to explain. Second, effect sizes vary strongly with design characteristics; there is no latent “true score” effect size at a theoretical level. Third, the practical impact of an effect may fade out or accumulate over time. Together, these arguments show that the magnitude of theory-testing effects does not provide reliable information about the magnitude of real-world effects. I conclude with recommendations for interpreting and reporting effect sizes in theory-testing research, emphasizing their utility for cumulative psychological science while cautioning against their uncritical interpretation as indicators of practical relevance.
"This article cautions against overinterpreting effect sizes in theory-testing psychology as indicators of practical relevance."
Open Access: doi.org/10.1525/coll...
🧵
New paper from me at Perspectives on Psychological Science!
"Reframing the Performance and Ethics of Empathic AI: Wisdom of the Crowd and Placebos"
I use analogies to two classic psychological effects to recast recent findings about the performance of empathy by LLMs.
doi.org/10.1177/1745...
Ever thought we acquire generalizable knowledge by discarding details and compressing our experiences?
In a new BBS paper, @sabinasloman.bsky.social and I argue otherwise, proposing a novel way of studying human learning inspired by double descent in ML.
Disagree? Propose a commentary by May 15 :)
Excited to be at #SANS2026! Looking forward to talking about information seeking about threat and reuniting with lovely colleagues. Come say hi!
What makes adaptive decision-making possible?
New preprint connecting cognitive, social, biological, and computational views! w/ Dorst, FeldmanHall, @smfleming.bsky.social, @catehartley.bsky.social, Gottlieb, Lejarraga, Müller-Trede, @angelaradulescu.bsky.social, and Rosati
osf.io/preprints/ps...
#SANS2026 thanks for being a great audience at my talk at this morning's symposium! check out our paper for more details.
If anyone is looking for a lab manager or RA (full or part-time), let me know. One of the most meticulous and detailed people that I know is looking for a position. They did a M.A. in psych with me here at Cornell.
We're hiring! Looking for a postdoc to work at UNSW Sydney, studying impacts of reward and information on attention, using eye-tracking, EEG, and modelling - with Kelly Garner, Daniel Pearson and me. Application link below, please spread the word!
external-careers.jobs.unsw.edu.au/cw/en/job/53...
P.S. There were so many people that were instrumental to this project and so much work had to come before this. I really cannot overstate how this paper was the result of a truly collective effort.
Moreover, when we checked back in with participants two months later, we found that participants whose rEC tracked Katz more strongly had a more cohesive friend group, an important source of social capital.
Possessing these neural maps also afforded key advantages for social navigation. Pattern differentiation of network members supported participants' judgments about how information might spread across the network in a separate task and this was modulated by how strongly the neural code tracked Katz.
As we hypothesized, we also find that these model-estimated maps of multistep connectivity are encoded by participants in the MTL, specifically the right entorhinal cortex (rEC). Spontaneous neural activity in the rEC differentiated network members who shared robust connections within the network.
This model (Katz) stitches together a more complete representation of the network by inferring likely but unobserved ties between network members connected by mutual connections, capturing participants' knowledge about their peers' friendships and the errors they tended to make.
By developing and comparing a series of computational models based on minimal assumptions about constraints on people's ability to acquire knowledge about others' friendships, we find evidence that people's mental representations of their network emerge from a process of multi-step abstraction.
To answer this question, we tracked the emerging & evolving real-world social network amongst incoming freshmen living across three dorms at Brown using social network surveys. We then probed their mental representations of the network by testing their knowledge of friendships amongst their peers.
However, we don't know how the brain represents people's real-world social networks. Borrowing from decades of work on physical navigation, we hypothesized that the medial temporal lobe (MTL), long implicated in representations of spatial relations, might also support maps of social networks.
We live our lives embedded in a complex web of relationships that comprise our social networks. Across different social interactions from strategic gossip to professional networking, knowing how others are connected within one's community is highly advantageous for navigating these interactions.
Now out in PNAS with @jaeyoungson.bsky.social, Alice Xia, @apaxon.bsky.social & @orielf.bsky.social. Medial temporal lobe encodes predictive representations of people's real-world social networks which afford them key advantages in social navigation. www.pnas.org/doi/10.1073/... 🧵
New preprint from my lab! We study how reinforcement learning & selective attention interact. To do so, we built a set of models describing different ways that value & reward prediction error can modulate top-down attention. We compare model outcomes to monkey data from a color value learning task
This is so, so well-articulated.
TBF this is a trend that has been going on for quite some time. Statistical software has become so user friendly that you can do some rather complex analyses these days without truly understanding your model
Is this a net bad? Hard to say. It’s on users to do their due diligence IMO
Part 2 of my shrinkage estimator series is out! Part 1 covered the univariate case, but now we dive into multivariate shrinkage 🤓
We cover Spearman's classic correlation disattenuation formula, multivariate James-Stein estimators, and hierarchical methods too
haines-lab.com/post/how-to-...