I'm honored to serve as an Expert Advisor for "The Alignment Project", a new initiative dedicated to ensuring AI systems are safe and beneficial. They are providing significant funding, compute, and collaboration opportunities for researchers---including cognitive scientists!.

RT @HuaDongXiong: Check out our 3 works with @Ji_An_Li @marcelomattar @NRDlab at #CogSci2025. All three works model learning to learn (meta….

RT @AISecurityInst: 📢Introducing the Alignment Project: A new fund for research on urgent challenges in AI alignment and control, backed by….

RT @geoffreyirving: I am very excited that AISI is announcing over £15M in funding for AI alignment and control, in partnership with other….

RT @smickdougle: Happy to announce that my lab @ Yale Psychology ( will be accepting PhD apps this year (for start….

RT @BasisOrg: We’re proud to announce the launch of AutumnBench, an open-source benchmark developed on our Autumn platform. This benchmark,….

RT @Ikuperwajs: New review on computational approaches to studying human planning out now in @TrendsCognSci! Really enjoyed having the oppo….

Here's the link to our paper: Here's the link to the Centaur paper: And here's Marcel Binz' thread on Centaur:

Also thrilled that this work appears in the same @nature issue as another paper I contributed to, Centaur! Like TinyRNNs, it excels at predicting human behavior. Impressively, Centaur works across many tasks, but this comes with a trade-off in model interpretability.

Crucially, we can discover these patterns for each individual! This is a game-changer for computational psychiatry, as it lets us pinpoint exactly how cognitive processes differ across people, without the constraints of a prespecified learning/decision model like RL.

This approach uncovered several patterns in the animal's decisions that previous models missed. Eg we found state-dependent learning rates, novel patterns of perseveration, and a peculiar type of forgetting where an action value decays towards the value of the alternative action.

But prediction is only half the story; we also need interpretability! We viewed tiny RNNs as dynamical systems with inputs (observations/rewards) and outputs (actions). Given their small size, we could visualize how their states evolved and discover the strategies they learned.

Across six different reward-learning tasks, tiny RNNs consistently outperform dozens of classical cognitive models in predicting the choices of individual animals and humans. Surprisingly, networks with just 2-4 units often performed the best in those simple lab tasks.

Our solution was to use very small RNNs, composed of 1-4 units. Those models are still great at modeling biological behavior without the need for pre-specified assumptions, but are small enough for us to interpret their mechanisms, combining the best of both worlds.

Cognitive modeling has always faced a dilemma. Traditional approaches based on normative or mechanistic assumptions are simple to understand but often cannot capture behavioral complexity. Large AI models are excellent predictors but are "black boxes" that are hard to interpret.

Thrilled to see our TinyRNN paper in @nature! We show how tiny RNNs predict choices of individual subjects accurately while staying fully interpretable. This approach can transform how we model cognitive processes in both healthy and disordered decisions.

A really fun project led by @Ji_An_Li adapting the neurofeedback methodology to LLMs!.

RT @SuryaGanguli: My @TEDAI2024 talk is out!.I discuss our work, spanning AI, physics, math & neuroscience, to deve….

What a cool opportunity!.

RT @patrickmineault: Excited to release what we’ve been working on at Amaranth Foundation, our latest whitepaper, NeuroAI for AI safety! A….