Akshay K. Jagadish
@akjagadish
Followers
348
Following
1K
Media
34
Statuses
692
Graduate student @CPILab. I study human and machine cognition.
Munich, Bavaria
Joined May 2013
1/ 🚨 Updated preprint:.“Generating Computational Cognitive Models using Large Language Models”. 👥 Co-led by @milenamr7 with Marvin Mathony, Tobias Ludwig & @cpilab. 📄 Check full paper here: .
5
10
27
RT @marcel_binz: Excited to see our Centaur project out in @Nature. TL;DR: Centaur is a computational model that predicts and simulates hum….
0
52
0
RT @a_proca: How do task dynamics impact learning in networks with internal dynamics?. Excited to share our ICML Oral paper on learning dyn….
0
18
0
RT @akjagadish: 1/ 🚨 Updated preprint:.“Generating Computational Cognitive Models using Large Language Models”. 👥 Co-led by @milenamr7 with….
0
10
0
RT @marcel_binz: New short-form preprint in which we use Centaur to identify gaps in interpretable cognitive models and revise them accordi….
0
4
0
Discovering cognitive models using LLMs has gained quite some traction in past few months/weeks. So, I want to highlight other concurrent works, as we are late to X party: .1. @pcastr and colleagues: 2. @JQ_Zhu and colleagues:
1/14 Can we build an AI that thinks like psychologists or economists? 🤔Our new preprint shows how reinforcement learning (RL) can train LLMs to explain human decisions—not just predict them! That is, we're pushing LLMs beyond mere prediction into explainable cognitive models.
0
0
1
10/ We are extremely grateful to all members of @cpilab for their feedback and support and to @MPICybernetics and @CompHealthMuc for funding! .
1/ 🚨 Updated preprint:.“Generating Computational Cognitive Models using Large Language Models”. 👥 Co-led by @milenamr7 with Marvin Mathony, Tobias Ludwig & @cpilab. 📄 Check full paper here: .
1
0
0
9/ 🌍 Implications:. We believe open source LLMs when powered by GeCCo pipeline have the potential to significantly advance cognitive modeling by democratizing access to complex model discovery and accelerating pace of research!.
1
0
0
8/ 🔬 Control experiments:.✔ Iterative feedback was most essential (removal worsened BIC).✔ No prompt contamination (LogProber: logB < 0.65).✔ Robust ground-truth recovery.✔ Matches/surpasses CENTAUR, which provides a proxy for explainable variance in human behavior
1
0
0
7/ 📊 Working memory:. LLM generated cognitive model surpassed the RL-WM baseline model from Rmus et al. 2023 . 🟢 Key strength: Captures adaptive shifts from fast WM-based learning at low load to slower RL-based learning under high load—mirroring human cognitive control.
1
0
0
6/ 📊 Planning:.LLM-generated cognitive models outperformed the hybrid RL baseline used in Feher da Silva et al. 2020. 🟢 Key strength: Uses separate learning rates for common vs. rare transitions, accurately modeling human stay/switch behavior in the two-step task.
1
0
0
5/ 📊 Learning:.LLM generated model exceeded the RW-4α baseline used by Chambon et al. 2020 . 🟢 Key strength: Introduces value decay and a fictive-value trace—closely reflecting frontopolar and striatal signals linked to counterfactual learning.
2
0
1
4/ 📊 Decision-making:.LLM-generated cognitive model outperformed the probabilistic weighted additive heuristic (pWADD) baseline in Hilbig et al. 2014. 🟢 strength: A single discount parameter adaptively spans EQW, TTB, and WADD—capturing strategy flexibility in human choices.
1
0
0
3/ ⚙️ GeCCo pipeline:.Prompt = task + behavioral data + guardrails + code template.→ LLM generates 3 candidate models.→ Models are fit offline using Bayesian Information Criterion (BIC).→ Best model fed back → 10 in-context refinement steps
1
0
1
2/ 💡 Motivation:.Cognitive models formalize human cognition but are often hand-crafted—slow, biased, and hard to scale. We introduce GeCCo: a pipeline using LLMs to generate interpretable models as Python functions, enabling fast, automated theory-building.
1
0
0
Please apply to join us! It’s a great lab to do science 🤗.
🚨 We're hiring! If you're excited about 🤖 ML/LLMs, 🧠 cognitive science, or 💭 computational psychiatry, come join us in Munich. Two fully funded PhDs @HelmholtzMunich: great mentorship, international vibe, lots of room to grow. 📅 May 16th.🔗
0
0
1
RT @marcel_binz: We are looking for two PhD students at our institute in Munich. Both postions are open-topic, so anything between cogniti….
0
80
0
RT @ZivBenZion1: Our study on AI chatbots mimicking human anxiety was featured on @Marketplace Tech, one of the most-listened-to shows in p….
0
6
0
Great to see that our work was covered by @nytimes! I still cannot believe it 😱.
🚨 Big News! Our research has been featured in The New York Times @nytimes! 🥳. This engagement highlights the real-world impact of science beyond academia and its role in shaping conversations on AI & mental health. 🔗 Read here:
0
0
2
RT @ZivBenZion1: 🚨 Big News! Our research has been featured in The New York Times @nytimes! 🥳. This engagement highlights the real-world im….
0
4
0