iiinova — where your digital self keeps evolving

Nathaniel Daw @nathanieldaw.bsky.social · 23/08/2026

You probably also never thought you'd hear me called "thoughtful" but I guess you just don't know me as well as my best friend gemini does.

070

Nathaniel Daw @nathanieldaw.bsky.social · 22/08/2026

life is full of haters but at least gemini thinks my plan to make my quiche vegetarian by replacing bacon with MSG + liquid smoke is "incredibly thoughtful."

4291

Nathaniel Daw @nathanieldaw.bsky.social · 14/07/2026

Provocative title ftw. (Tho it inspired @kristorpjensen.bsky.social to give a talk titled "planning in the brain: it's not what nathaniel thinks it is")

0194

Reposted by Nathaniel Daw

Akshay K. Jagadish @akjagadish.bsky.social · 26/06/2026

1/ 🚨 New preprint: "Closing the Loop to Discover Psychological Theories with an Automated Cognitive Scientist" Introducing AutoCog 🤖 — a fully autonomous AI system that runs the entire scientific discovery cycle in cognitive science to surface novel theories of human behavior 🧵

1185

Reposted by Nathaniel Daw

Laura Luyten @lauraluyten.bsky.social · 21/06/2026

I've been a bit quiet lately (lots of teaching duties...), but I am glad to share that I will have some academic "me time" for the next few weeks. I will be at the Princeton Neuroscience Institute & Department of Psychology, for a research stay hosted by @nathanieldaw.bsky.social!

0132

Reposted by Nathaniel Daw

Noémi Éltető @noemielteto.bsky.social · 19/06/2026

First paper since joining Google DeepMind! We present 🌍ATLAS (Active Theory Learning for Automated Science), a pipeline that generates interpretable mechanistic models from data and optimizes experiments to test them. Thread below

1377

Nathaniel Daw @nathanieldaw.bsky.social · 07/06/2026

I don't think the behavior is so well adapted to this task instance. Noise is not actually so low: high learning rate is a strange observation which I think demands explanation. I think it's a more general strategy (though again very much agree this needs exploring) & have many longer thoughts why.

020

Nathaniel Daw @nathanieldaw.bsky.social · 07/06/2026

This project convinced me it's "easy" to get a computer to run up a score, harder to define the score you actually want to run up. I think we moved the ball by adding the 2nd objective but I agree we could go a lot farther, if we could score subjective things well. (We have used LLM judge some.)

010

Nathaniel Daw @nathanieldaw.bsky.social · 05/06/2026

Also it's hard to overemphasize the role of RPE mechanisms driving the task/model space historically. This task was explicitly designed (in 2006) for the kalman filter. Contra your point, the discovered model innovations here are very much not that.

120

Nathaniel Daw @nathanieldaw.bsky.social · 05/06/2026

I increasingly think (like you) big question then is how ppl adapt to tasks, rather than within. This study didn't sample across tasks, clear next step. But I think the results point that way. Eg the human reward rule (slow aspiration level ratchet) may be outer-loop adaptation to task statistics

140

Nathaniel Daw @nathanieldaw.bsky.social · 05/06/2026

Too much to say for tweets! I think model building in this area has had the opposite aesthetic in practice - fitting models to tasks (& vice versa, within a class of tasks amenable to incremental running avg). Bayes-ish rationale: the informed or asymptotic ideal observer should be task optimized 1/

110

Nathaniel Daw @nathanieldaw.bsky.social · 04/06/2026

Lotta hallucination in this figure eh?

020

Nathaniel Daw @nathanieldaw.bsky.social · 03/06/2026

I have been so excited about this project. In many ways the most surprising thing to me was how seemingly similar the AI-discovered models were to simple ones I thought I understood, even though they fit way better. This means we can also really get what makes them work.

1438

Nathaniel Daw @nathanieldaw.bsky.social · 22/04/2026

If I understand you, I think you might actually be receptive to our argument which is basically that the classic AI-inspired notion of "planning" does not correspond to a discrete thing in the brain, and the notion that it does is, I suppose, a failure of "are for" thinking.

030

Nathaniel Daw @nathanieldaw.bsky.social · 11/02/2026

Thanks for the kind words and also the cogent summary. This was such a challenging study to wrap our heads around & I think your tweet is better than our abstract.

020

Nathaniel Daw @nathanieldaw.bsky.social · 14/12/2025

Reminds me of how Google maps, some years ago, reached 3.8 miles away to call my attention to this establishment of interest

130

Nathaniel Daw @nathanieldaw.bsky.social · 13/12/2025

With all respect (really!) it would make a great 3000-word TICS position piece: it's funny and thought provoking. But stretching it out to book length would be bonkers; I can't see either the joke or the serious point lasting.

010

Reposted by Nathaniel Daw

Alice Zhang @licezhang.bsky.social · 15/10/2025

New paper out in cognition with @arikahn.bsky.social, @nathanieldaw.bsky.social, Cate Hartley, and @katenuss.bsky.social !! We show that children 👶 use predictive representations (e.g. SR) to guide their choices, providing an account of how they can make flexible choices in a changing world

sciencedirect.com

Children leverage predictive representations for flexible, value-guided choice

By harnessing a mental model of how the world works, learners can make flexible choices in changing environments. However, while children and adolesce…

14713

Nathaniel Daw @nathanieldaw.bsky.social · 15/10/2025

This is also why Sacks' writing is so extraordinary.

1170

Nathaniel Daw @nathanieldaw.bsky.social · 14/10/2025

They didn't need the scientific literature, nor faculty input, to do the right thing all along. They have always had more access to private data, which must tell the same story as the public data even more clearly. It's distressing to admit but the only thing that has changed is the political winds.

010

Nathaniel Daw @nathanieldaw.bsky.social · 13/10/2025

I don't think anyone realized anything. They had to know all along if you and I did: just different factors weighed differently at different times.

110

Nathaniel Daw @nathanieldaw.bsky.social · 13/10/2025

It's embarrassing and outrageous it took us five years to bring them back.

130

Nathaniel Daw @nathanieldaw.bsky.social · 10/10/2025

This is the best news for Princeton since John Nash's thesis. admission.princeton.edu/apply/standa...

admission.princeton.edu

Standardized Testing | Princeton Admission

080

Nathaniel Daw @nathanieldaw.bsky.social · 20/09/2025

I think the Nazi submarine at science and industry museum is extraordinary, as is pequod's pizza.

010

Nathaniel Daw @nathanieldaw.bsky.social · 20/09/2025

(I'm actually just trying to bait @nicolecrust.bsky.social into scolding you for my reactionary tweets again.)

130

Nathaniel Daw @nathanieldaw.bsky.social · 20/09/2025

Do you think Lord Krebs would have let trainees speak let alone squeal with delight?

280

Nathaniel Daw @nathanieldaw.bsky.social · 20/09/2025

Excited to read this paper but you should really clamp down on this undecorous emotion in your lab meeting.

1100

Nathaniel Daw @nathanieldaw.bsky.social · 15/09/2025

I am not sure how far this goes back in Earl's thinking. For my own part I legit expected Earl's early experiments to find grandmother cells for task set variables in pfc (maybe based on my own shallow reading of eg miller & cohen) and was surprised when he found mixed codes instead.

060

Nathaniel Daw @nathanieldaw.bsky.social · 15/09/2025

To be fair, Earl was probably fifteen years ahead of this becoming a commonplace idea in deep networks (which arguably happened only quite recently with the ideas about superposition in LLMs)

150

Reposted by Nathaniel Daw

Norman Lab @ptoncompmemlab.bsky.social · 13/08/2025

Come work with us! @princetonneuro.bsky.social and the Department of Psychology at Princeton University are searching for a tenure-track Assistant Professor in the area of human cognitive neuroscience, to be hired jointly in Psychology and Neuroscience: puwebp.princeton.edu/AcadHire/app...

puwebp.princeton.edu

27063

Reposted by Nathaniel Daw

Payam Piray @payampiray.bsky.social · 12/08/2025

New paper with @nathanieldaw.bsky.social in Nature Communications: an RL model that builds a successor map compositionally. The new model plans as well as the best models, and it links components of the map used for planning to neural codes in the medial entorhinal cortex. rdcu.be/eAofi

rdcu.be

Reconciling flexibility and efficiency: medial entorhinal cortex represents a compositional cognitive map

Nature Communications - How the brain creates compositional cognitive maps that support both flexible and efficient planning remains poorly understood. Here, authors propose a...

17820

Nathaniel Daw @nathanieldaw.bsky.social · 23/07/2025

Not exactly related but not exactly unrelated either: xkcd.com/793

xkcd.com

Physicists

140

Nathaniel Daw @nathanieldaw.bsky.social · 22/07/2025

Have you considered that maybe your problem is that you are talking to physicists?

160

Reposted by Nathaniel Daw

Claire Gillan @clairegillan.bsky.social · 01/07/2025

We @smfleming.bsky.social, Marion Rouault and @seowxft.bsky.social and I) have posted a reply osf.io/preprints/ps... to a preprint that recently raised concerns about the validity of associations between mental health and metacognition from online studies. I hope you can take the time to read it.

osf.io

OSF

23721

Nathaniel Daw @nathanieldaw.bsky.social · 21/05/2025

grumpy-old-men-muppet-show.gif that I don't know how to send.

060

Nathaniel Daw @nathanieldaw.bsky.social · 21/05/2025

I don't think we disagree. 🤷‍♂️

010

Nathaniel Daw @nathanieldaw.bsky.social · 20/05/2025

Also I constantly remind my coauthors that it is easily verified that ~nobody expands the thread so the first tweet & link are the whole game. Write a good tweet, leading to a good abstract leading to a good paper. Spare me the emojis.

381

Nathaniel Daw @nathanieldaw.bsky.social · 20/05/2025

I like organizing project history in a deep nested tree of folders labeled "new," "old," "newer," and "older"

160

Nathaniel Daw @nathanieldaw.bsky.social · 11/04/2025

heads up on a cool postdoc at cambridge. i collaborate on this project as well and looking forward to interacting with the team.

081

Nathaniel Daw @nathanieldaw.bsky.social · 09/04/2025

i know of one other almost identical case.

010

Nathaniel Daw @nathanieldaw.bsky.social · 03/04/2025

Right I should have said most of those recs ultimately came from Nina. I second the civil rights museum too.

130

Nathaniel Daw @nathanieldaw.bsky.social · 03/04/2025

Go to bomb biscuit and fishmonger; also walk on the beltline. Oh and I haven't been but people say delbar is great persian

130

Nathaniel Daw @nathanieldaw.bsky.social · 30/03/2025

It is indeed truly a fearful thing to agree with me

080