Ariel Goldstein

Do LLMs have motivation? Motivation is a key lens for explaining human behavior. As LLM behavior becomes more human-like, a natural question arises: could it help understand model behavior too? With @AsaelSklar @GoldsteinYAriel @roireichart 📄 Paper: arxiv.org/pdf/2603.14347 1/5

I’m hiring a PhD student for my lab at Cambridge’s Center for Human Inspired AI (CHIA)! Work at the intersection of AI × cognition × neuroscience. 📅 Deadline: Dec 2 🔗 Program: postgraduate.study.cam.ac.uk/courses/direct… 📚 My Google Scholar: scholar.google.com/citations?user…

🚨 New preprint! One idea, many ways to say it – does your brain track those options before you speak? Using LLMs, we put this to the test: biorxiv.org/content/10.110… We show for the 1st time that the brain represents many alternatives simultaneously in both listening & speaking 🧵

Most novel insight: When an LLM recognizes a bias, it’s actually less likely to exhibit it. This helps explain conflicting results in prior studies—each was seeing a subset of the broader pattern.

New paper out! Do LLMs Exhibit Human-Like Cognitive Biases? We systematically tested 20 biases across 10 leading LLMs using pre-registered experiments. The result: cognitive biases in LLMs are both pervasive and nuanced. papers.ssrn.com/sol3/papers.cf…

New Preprint 🎉 LLM self-assessment unlocks efficient decoding ✅ Our Confidence-Informed Self-Consistency (CISC) method cuts compute without losing accuracy. We also rethink confidence evaluation & contribute to the debate on self-verification. arxiv.org/abs/2502.06233 1/8👇

Very excited to share our new paper published in Nature Communications @NatureComms (link below). This work is part of my PhD research under the supervision of @roireichart (Technion), @HassonUri (@HassonLab), and @ArielYGoldstein, in collaboration with @YoavMeiri.

@yoavgo Second, drawing on our human experience where sometimes (we feel) that humans who talk about an issue/problem can reach a conclusion that separately they couldn't have - worth mimicking no?

let's talk about "agents" (in the LLM sense). there's a lot of buzz around "multi-agent" systems where agents collaborate but... i don't really get how it differs from a thinking of a single agent with multiple modes of operation. what are the benefits of modeling as multi-agent?

@yoavgo First, if you are interested in verbal-based interactions, artificial agent systems serve as a "lab" for different hypotheses about language-based interactions.

@AsafYehudai @marcel_binz @huggingface All LLMs' embeddings (that I tested) correlate with brain. It is interesting to see if it explain more. However, predicting behavior using LLMs is more challenging IMO.

@marcel_binz @huggingface @GoldsteinYAriel What do you think about this? Worth comparing this model's representations with brain signals?

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/Lla…

🎯 Finally, we leverage these insights to introduce a new token-level early-exit strategy that beats existing methods in balancing performance and efficiency. More accurate predictions and faster models—win-win! Joint work with @GoldsteinYAriel @TomerSchlank @GabiStanovsky 4/4

We propose an underlying task transition mechanism where each task corresponds to determining the k-th ranking token. By understanding these transitions, we can predict the current task from hidden layers representations and cause the model to switch between tasks! 3/4

Our findings reveal that this is true across language, vision, and speech models and across architecture variants: decoder-only, encoder-only, and full-Transformer. It even occurs in untrained Transformer models! 🤯 2/4

📢Paper release📢 What computation is the Transformer performing in the layers after the top-1 becomes fixed (a so called "saturation event")? We show that the next highest-ranked tokens also undergo saturation *in order* of their ranking. Preprint: arxiv.org/abs/2410.20210 1/4

🧠🤖 How do LLMs think? What kind of thought processes can emerge from artificial intelligence? Our latest paper about multi-hop reasoning tasks reveals some new interesting insights. Check out this thread for more details! arxiv.org/abs/2406.13858 @GoldsteinYAriel @amir_feder

Chreck out our new LLM-based suicide research paper at @FrontPsychiatry !

🧠🤖 Does learning in the brain inherently require plasticity? In our latest paper we question this assumption, by leveraging insights into how LLMs "learn". Check out this thread for more details! biorxiv.org/content/10.110… w/ @YuvalShalev1 @GabiStanovsky @GoldsteinYAriel