Sabitlenmiş Tweet
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:
arxiv.org/abs/2502.00879
English
Akshay K. Jagadish
720 posts
Akshay K. Jagadish
@akjagadish
Research Fellow, @Princeton AI Lab. I use AI to study natural and artificial minds. PhD @CPILab @MPICybernetics @Uni_Tue
Princeton, NJ Katılım Mayıs 2013
595 Takip Edilen391 Takipçiler
Akshay K. Jagadish retweetledi
How is uncertainty in LLMs output reflected in internal representations?
In our new work (to appear at ICML 2026), we show that the shape of internal token trajectories provides a direct geometric link to behavioral uncertainty (output entropy). 🧵(1/n)
English
Every project needs a @milenamr7 ;)
Mayank Agrawal@mayankagrawal
Meet Milena Rmus (@milenamr7) — engineering wizard, data visualization artist, and resident cat/meme/ClaudeCode extraordinaire at @RoundtableHQ_ . Every team needs a Milena. Her love of the craft shows up everywhere: building elegant visualizations for exploratory statistical work, monitoring API errors from the middle of metal concerts, and keeping company culture alive with perfectly timed memes. She’s a driving force behind Roundtable’s Proof of Human research agenda. We met through overlapping computational cognitive science circles — she holds a BA from Brown, a PhD from Anne Collins’ lab at UC Berkeley, and most recently completed a postdoc in Munich. Her publication record is extensive: Nature, NeurIPS, PLOS Computational Biology. Again and again, she brings cutting-edge AI and ML approaches to understanding the human mind and brain. At Roundtable, she leads continuous benchmarking and red-teaming of the Proof of Human API. Fraud evolves quickly — having our toughest critics in-house is a feature, not a bug. And don’t underestimate her intellectual edge. She led our “AI Capabilities ≠ Humanness” piece, grounding a contrarian AI stance in cognitive science. More work is on the way, and we’re excited to share it soon :) In the meantime — to honor Milena — here are a few cat memes.
English
Akshay K. Jagadish retweetledi
Meet Milena Rmus (@milenamr7) — engineering wizard, data visualization artist, and resident cat/meme/ClaudeCode extraordinaire at @RoundtableHQ_ .
Every team needs a Milena. Her love of the craft shows up everywhere: building elegant visualizations for exploratory statistical work, monitoring API errors from the middle of metal concerts, and keeping company culture alive with perfectly timed memes.
She’s a driving force behind Roundtable’s Proof of Human research agenda. We met through overlapping computational cognitive science circles — she holds a BA from Brown, a PhD from Anne Collins’ lab at UC Berkeley, and most recently completed a postdoc in Munich.
Her publication record is extensive: Nature, NeurIPS, PLOS Computational Biology. Again and again, she brings cutting-edge AI and ML approaches to understanding the human mind and brain.
At Roundtable, she leads continuous benchmarking and red-teaming of the Proof of Human API. Fraud evolves quickly — having our toughest critics in-house is a feature, not a bug.
And don’t underestimate her intellectual edge. She led our “AI Capabilities ≠ Humanness” piece, grounding a contrarian AI stance in cognitive science. More work is on the way, and we’re excited to share it soon :)
In the meantime — to honor Milena — here are a few cat memes.
English
This is crucial ingredient required for automating scientific discovery :)
Joy He-Yueya@JoyHeYueya
Scientists often make breakthroughs by synthesizing ideas across papers. In our new paper, we ask whether a language model can anticipate this process: given two parent papers, can it generate the core insight of a future paper built on them? 🧵⬇️
English
I was going to build my founder OS in Notion. The tinkerer won.
One weekend in Claude Code: calendar + email + Slack + finance + CRM, one dashboard, natural-language queryable end to end.
English
📘 Excited to share that Decision Making: A Very Short Introduction (OUP) is now available online (PDF for subscribers):
academic.oup.com/book/62545?log…
What is it about? Understanding how humans (and other agents) make choices. Below a bit more information👇
English
This is cool!
Mingchen Zhuge@MingchenZhuge
🫱 Introducing 𝐍𝐞𝐮𝐫𝐚𝐥 𝐂𝐨𝐦𝐩𝐮𝐭𝐞𝐫s: 𝐰𝐡𝐚𝐭 𝐢𝐟 𝐀𝐈 𝐝𝐨𝐞𝐬 𝐧𝐨𝐭 𝐣𝐮𝐬𝐭 𝐮𝐬𝐞 𝐜𝐨𝐦𝐩𝐮𝐭𝐞𝐫𝐬 𝐛𝐞𝐭𝐭𝐞𝐫, 𝐛𝐮𝐭 𝐛𝐞𝐠𝐢𝐧𝐬 𝐭𝐨 𝐛𝐞𝐜𝐨𝐦𝐞 𝐭𝐡𝐞 𝐫𝐮𝐧𝐧𝐢𝐧𝐠 𝐜𝐨𝐦𝐩𝐮𝐭𝐞𝐫 𝐢𝐭𝐬𝐞𝐥𝐟? Beyond today's conventional computers, agents, and world models, Neural Computers (NCs) are new frontiers where computation, memory, and I/O move into a learned runtime state. We ask: whether parts of runtime can move inward into the learning system itself. This is our first step toward the Completely Neural Computer (CNC): a general-purpose neural computer with stable execution, explicit reprogramming, and durable capability reuse. Work done with Mingchen Zhuge (@MingchenZhuge), Changsheng Zhao, Haozhe Liu (@HaoZhe65347 ), Zijian Zhou (@ZijianZhou524 ), Shuming Liu (@shuming96 ), Wenyi Wang (@Wenyi_AI_Wang ), Ernie Chang (@erniecyc ), Gael Le Lan, Junjie Fei, Wenxuan Zhang, Zhipeng Cai (@cai_zhipeng ), Zechun Liu (@zechunliu ), Yunyang Xiong (@YoungXiong1 ), Yining Yang, Yuandong Tian (@tydsh ), Yangyang Shi, Vikas Chandra (@vikasc), Juergen Schmidhuber (@SchmidhuberAI)
English
Akshay K. Jagadish retweetledi
🫱 Introducing 𝐍𝐞𝐮𝐫𝐚𝐥 𝐂𝐨𝐦𝐩𝐮𝐭𝐞𝐫s:
𝐰𝐡𝐚𝐭 𝐢𝐟 𝐀𝐈 𝐝𝐨𝐞𝐬 𝐧𝐨𝐭 𝐣𝐮𝐬𝐭 𝐮𝐬𝐞 𝐜𝐨𝐦𝐩𝐮𝐭𝐞𝐫𝐬 𝐛𝐞𝐭𝐭𝐞𝐫, 𝐛𝐮𝐭 𝐛𝐞𝐠𝐢𝐧𝐬 𝐭𝐨 𝐛𝐞𝐜𝐨𝐦𝐞 𝐭𝐡𝐞 𝐫𝐮𝐧𝐧𝐢𝐧𝐠 𝐜𝐨𝐦𝐩𝐮𝐭𝐞𝐫 𝐢𝐭𝐬𝐞𝐥𝐟?
Beyond today's conventional computers, agents, and world models, Neural Computers (NCs) are new frontiers where computation, memory, and I/O move into a learned runtime state.
We ask: whether parts of runtime can move inward into the learning system itself. This is our first step toward the Completely Neural Computer (CNC): a general-purpose neural computer with stable execution, explicit reprogramming, and durable capability reuse.
Work done with Mingchen Zhuge (@MingchenZhuge), Changsheng Zhao, Haozhe Liu (@HaoZhe65347 ), Zijian Zhou (@ZijianZhou524 ), Shuming Liu (@shuming96 ), Wenyi Wang (@Wenyi_AI_Wang ), Ernie Chang (@erniecyc ), Gael Le Lan, Junjie Fei, Wenxuan Zhang, Zhipeng Cai (@cai_zhipeng ), Zechun Liu (@zechunliu ), Yunyang Xiong (@YoungXiong1 ), Yining Yang, Yuandong Tian (@tydsh ), Yangyang Shi, Vikas Chandra (@vikasc), Juergen Schmidhuber (@SchmidhuberAI)
English
Akshay K. Jagadish retweetledi
I am thrilled to announce that I will be joining @Cambridge_Uni as Assistant Professor in Machine Learning at @Cambridge_Eng starting January 2027!
English
Akshay K. Jagadish retweetledi
Today we present a new framework for measuring human-like general intelligence in machines (what some people call AGI).
Conventional AI benchmarks today assess only narrow capabilities in a limited range of human activities.
We propose that a more promising way to evaluate human-like general intelligence in AI systems is through a particularly strong form of general game playing: studying how and how well they play and learn to play all conceivable human games — what we call the ``Multiverse of Human Games''.
Taking a first step towards this vision, we introduce the AI GameStore, a scalable and open-ended platform that uses LLMs with humans-in-the-loop to automatically construct standardized and containerized variants of popular human games on digital gaming platforms.
As a proof of concept, we generated 100 such games based on the top charts of Apple App Store and Steam, and evaluated seven frontier vision-language models (VLMs) on short episodes of play. The best models achieved less than 10% of the human average score on the majority of the games.
Check out our website to play the games, see how agents play, and build agents to solve them!
English
Akshay K. Jagadish retweetledi
Why don’t neural networks learn all at once, but instead progress from simple to complex solutions? And what does “simple” even mean across different neural network architectures?
Sharing our new paper @iclr_conf led by Yedi Zhang with Peter Latham
arxiv.org/abs/2512.20607
GIF
English
This work was accepted to @iclr_conf check out its latest version: arxiv.org/abs/2505.11614
Zhu Jian-Qiao@JQ_Zhu
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.
English
🚨 NEW PAPER: “POPE: Learning to Reason on Hard Problems via Privileged On-Policy Exploration”!
❓ How do we train LLMs with RL on hard problems when the model never gets a single correct rollout?
💡 Short answer: standard RL is stuck. We show why, and introduce POPE to break this deadlock.
🧵[1/N]
English
the battle between classic cognitive models and LLMs fine-tuned on human data
English
@kzkirie @agopal42 @HochreiterSepp I didn’t know about it. Sounds very relevant. Thanks for sharing! :)
English
Am I the only one who thinks Cotter & Conwell's "Fixed-weight networks can learn" (1990) is so underappreciated in light of in-context learning (and beyond)? with follow-ups including @HochreiterSepp's masterpiece (2001).
Cited 62 times (of which 10% are me). Seems too few.
English
🚨 @milenamr7 and I will be presenting this work at AI4science workshop at Neurips-2025 today!
Time: 11.20 am - 12.30 pm
Location: Room 20, upper level
Workshop: AI4Science
Please drop if you are interested in iterative program synthesis and automated cognitive modeling!
Marcel Binz@marcel_binz
New short-form preprint in which we use Centaur to identify gaps in interpretable cognitive models and revise them accordingly using Qwen3 -- fully automated and without a human-in-the-loop. arxiv.org/abs/2505.17661
English
Akshay K. Jagadish retweetledi
Drop by and see our first step towards closing the loop for automating discovery in cognitive science. Much more to come on this soon, so talk to Akshay and Milena about it.
Akshay K. Jagadish@akjagadish
🚨 @milenamr7 and I will be presenting this work at AI4science workshop at Neurips-2025 today! Time: 11.20 am - 12.30 pm Location: Room 20, upper level Workshop: AI4Science Please drop if you are interested in iterative program synthesis and automated cognitive modeling!
English
Thanks for the open invite, @pcastr! No better way to enjoy the SD morning than a run along the coast :)
Pablo Samuel Castro@pcastr
Third #runconference at #NeurIPS2025 was great, including special guest @JeffDean ! Same place, same time, tomorrow morning if you want to join! 🤖🏃🏾
English