Computational Cognitive Scientist, Center for Cognitive and Decision Sciences, University of Basel.
Basel, Switzerland
Joined April 2020
- Tweets35
- Following78
- Followers90
- Likes78
Zak Hussain retweeted
🚨 New publication 🚨
Excited to see this published in Findings of #ACL2025, led by @ZakASHussain.
We critically evaluate claims that #LLMs are "just next token predictors" or "just machines" and call for a more measured discussion on LLM cognition.
aclanthology.org/2025.findin…
Zak Hussain retweeted
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.
Zak Hussain retweeted
Happy to have concluded another iteration of our 5-day open LLMs course, together with @ZakASHussain.
If you are interested in LLMs for behavioral and social sciences, check out our...
Tutorial: doi.org/10.3758/s13428-024-0…
Open materials: github.com/zak-Hussain/LLM4B…
Zak Hussain retweeted
How can we reduce conceptual clutter in the psychological sciences?
@rui__mata and I propose a solution based on a fine-tuned 🤖 LLM (bit.ly/mpnet-pers) and test it for 🎭 personality psychology.
The paper is finally out in @NatureHumBehav: go.nature.com/4bEaaja
Zak Hussain retweeted
Excited to share the materials for our course on open LLMs for science of science research @ZakASHussain
and I offered at the recent meeting of @euroscisci
at @LMU_Muenchen.
github.com/zak-Hussain/LLM4S…
Zak Hussain retweeted
Do LLMs think?
Excited to share our updated preprint critically discussing two "Justaic" stances claiming that LLMs lack cognition because they are "just" next-token predictors or "just" machines.
Led by @ZakASHussain and with @rui__mata.
osf.io/preprints/osf/y34ur_v…
Zak Hussain retweeted
The potential of LLMs in social & behavioral science is enormous—but how can we leverage them?
@ZakASHussain & I just taught a 5-day course at #GSERM Ljubljana on this. Check out our open materials (cc-by-sa) on using open LLMs with @huggingface:
github.com/Zak-Hussain/LLM4B…
Zak Hussain retweeted
🚨 New preprint 🚨
Excited to share this preprint led by @ZakASHussain (with @rui__mata and Ben Newell), comparing the semantic contents of embedding models from text, brain, and behavior data.
We find that behavior captures key psychological dimensions better than text, which has implications for language model training and evaluation.
A thread.
youtube.com/watch?v=NvfK1TkX…
This was fun! Looking forward to seeing more research on continual learning (in addition to the "transient" status quo)
Zak Hussain retweeted
Excited to announce Centaur -- the first foundation model of human cognition. Centaur can predict and simulate human behavior in any experiment expressible in natural language. You can readily download the model from @huggingface and test it yourself: huggingface.co/marcelbinz/Ll…
Zak Hussain retweeted
Preprint: Individual-level semantic networks are a critical ingredient for building realistic cognitive models. In our simulation, @rui__mata, @dirkuwulff, and I show when measurements of individual semantic networks are accurate and when they are not.
arxiv.org/abs/2410.18326
arxiv.org/pdf/2408.05859
Cool to see the conceptual tools of cognitive science applied directly to challenges in artificial neural network interpretability - especially the distinction between *semantic* and *algorithmic* interpretation.
Zak Hussain retweeted
New preprint in which use methods from mech interp to reveal that Llama-3-70B implements TD-learning in-context.
Can LLMs do reinforcement learning in-context - and if so, how do they do it? Using Sparse Autoencoders, we find that Llama 3 relies on representations resembling TD errors, Q-values and even the SR to learn in three RL tasks in-context! Co-lead with the inimitable @can_demircann
youtube.com/watch?v=meUtWtY0…
Really impressed by the clarity of Solm's explanations here. Also cool to hear a new (at least to me), affect-centered perspective on consciousness.
Zak Hussain retweeted
Great to see this perspective led by @jason_w_burton finally out in @NatureHumBehav.
Article: nature.com/articles/s41562-0…
How can we make the best possible use of large language models (LLMs) for a smarter and more inclusive society? New article @NatureHumBehav outlines the ways LLMs can help and hurt collective intelligence and proposes recommendations for action. mpib-berlin.mpg.de/press-rel…
Zak Hussain retweeted
Important antitrust lawsuit against academic publishers.
We don't have to accept the status quo, in which these companies make huge profits on the backs of scholars, and damage science in the process.
lieffcabraser.com/antitrust/…
Zak Hussain retweeted
🚨 New preprint 🚨
I am excited to share a piece with @rui__mata and @ZakASHussain in which we advocate for a greater reliance on open LLMs in the behavioral and social sciences. Despite the availability of powerful, more reproducible open alternatives (e.g., via @huggingface), these fields continue to favor closed models.
Preprint: osf.io/preprints/osf/ybvzs
Zak Hussain retweeted
🚨 New article 🚨
Open LLMs are powerful and reproducible alternatives to closed models like GPT or Gemini.
In this tutorial, we (@ZakASHussain, @marcel_binz, @rui__mata) explain how LLMs work and show how to apply open LLMs to behavioral science questions using the @huggingface ecosystem.
Article: link.springer.com/article/10…
Repository: github.com/Zak-Hussain/LLM4B…
Zak Hussain retweeted
🚨Call for help🚨
For our piece 'Are LLMs "just" next-token predictors?" (osf.io/preprints/osf/y34ur), we (@ZakASHussain, @rui__mata) are looking for quotes of unsubstantiated deflationary claims about LLM cognition in science or media.
See github.com/Zak-Hussain/again….
🙏Thank you!!
Zak Hussain retweeted
[Review] "Modern language models and vector-symbolic architectures show that vector-based models are capable of handling the compositional, structured, and symbolic properties required for human concepts" sciencedirect.com/science/ar…
