iiinova — where your digital self keeps evolving

Reposted by Francesco Poli

Cian O'Donnell @cianodonnell.bsky.social · 07/10/2026

The field of mathematics is being devastated in front of our eyes. For neuroscience, experimentalists may be safe, but how long until the AI-research wave takes over our computational/theoretical neuroscience? Is there any reason they can't? They can code, they can do maths. They know neurobiology

7223

Francesco Poli @francescopoli.bsky.social · 07/10/2026

Thanks so much, a simple RL model might indeed work well here. I will probably paly around trying to generate datasets where participants seem similar in their performance but some have high learning rate + low temp and other have low learning rate + high temp.

010

Francesco Poli @francescopoli.bsky.social · 07/10/2026

thanks so much, this looks super interesting (also research topic-wise) - will definitely take a look!

010

Francesco Poli @francescopoli.bsky.social · 06/10/2026

HELP! What is your go-to example to show the importance of computational modelling in cognitive science? I will give an intro workshop on this and I'm looking for an example that clearly and simply illustrates why sometimes modelling is better than looking just at the raw data...

251

Francesco Poli @francescopoli.bsky.social · 04/10/2026

Yes! I guess I'm just saying we need more of those kinds of models :)

010

Francesco Poli @francescopoli.bsky.social · 04/10/2026

2. Lots of cognitive (neuro)science has been so much about rational agents and optimal learning. This will be obsolete, and the challenge will be to make machine that struggle and fail like people

010

Francesco Poli @francescopoli.bsky.social · 04/10/2026

1. AI will soon be able to simulate this content as well, at which point we will only be left with the option to pick up a pen and do imperfect art ourselves, cutting out the "consuming content" bit and freed up by the burden of it needing to be perfect

100

Francesco Poli @francescopoli.bsky.social · 04/10/2026

A very interesting response to AI content is imperfect human content, where errors are a signature of "humanness". I think this is beautiful for to independent reasons:

222

Francesco Poli @francescopoli.bsky.social · 02/10/2026

So excited to share that I will be starting my research group at the Max Planck Institute for Evolutionary Anthropology! 🥳 My lab will study curiosity and exploration across biological (and artificial) minds. I will recruit a PhD, a postdoc and an RA in the coming months, so please spread the word!

86514

Francesco Poli @francescopoli.bsky.social · 27/09/2026

I can't overstate the importance of this toolbox & this paper! Really amazing stuff

031

Francesco Poli @francescopoli.bsky.social · 24/09/2026

My guess is that the most likely scenario is somewhere in between, with most people doing (the very much needed) computational work, and only a minority of people actually collecting high quality data. Unfortunately, the selection and publication systems are not gonna change any time soon

000

Francesco Poli @francescopoli.bsky.social · 24/09/2026

We will all hastily collect data with online cognitive tasks, which unbeknownst to us LLMs will be actually performing, and reanalyse existing data just because it's available.

100

Francesco Poli @francescopoli.bsky.social · 24/09/2026

The negative scenario is that computational research will go much faster than experimental research, and with today's selection criteria, computational folks will be endlessly more productive than experimenters. So actual experimental work will die out.

110

Francesco Poli @francescopoli.bsky.social · 24/09/2026

Experimental work will be the main bottleneck to scientific progress. And by virtue of that, it will become more important than ever. Well, that's the positive scenario.

110

Francesco Poli @francescopoli.bsky.social · 24/09/2026

The lab will always be the heart of scientific work. Every step of the scientific process, from reviewing the literature to analysing data and writing manuscripts will be faster than ever because of LLMs, EXCEPT experimental work. That takes time and painstaking effort, but...

240

Reposted by Francesco Poli

Philip N Cohen @philipncohen.com · 03/09/2026

Part 2. There is no bottom to Ariely's malfeasance. The Arday tragedy is so raw now the comparison is unavoidable. Ariely's fraud cost millions. Books were sold, careers altered. People and organizations changed policies and practices based on his fake results...

25613

Reposted by Francesco Poli

Mark Riedl @markriedl.bsky.social · 31/08/2026

Yann LeCun throwing bombs at the ACM AI Leadership Summit

2020423

Francesco Poli @francescopoli.bsky.social · 28/08/2026

Amazing project with amazing people in an amazing institute! Do reach out if you have any questions :)

030

Reposted by Francesco Poli

International Society of Developmental Biology @isdb.bsky.social · 18/08/2026

New preprint from the @mayorlab.bsky.social showing neural tube folding and neural crest migration are in a feedback loop 🔁. Neural tube folding compresses neural crest and triggers MMP14 expression, which drives their migration, which itself drives neural tube folding! doi.org/10.64898/202...

25213

Reposted by Francesco Poli

Christoph Völter @christophvoelter.bsky.social · 28/07/2026

Join us in Vienna! We are looking for a #PhD student and a #Postdoc to investigate #innovation in great #apes: PhD ad: share.eva.mpg.de/index.php/s/... Postdoc ad: share.eva.mpg.de/index.php/s/... Application deadline: August 17th. Please share widely!

A chimp solving a puzzle box problem. Photo credit: MPI-EVA

19083

Francesco Poli @francescopoli.bsky.social · 24/07/2026

I've always been puzzled by how a sample size of 2 would be completely unacceptable for empirical work, but the norm for peer-reviews... Is 2 significantly better than 0? Has this been tested?

071

Reposted by Francesco Poli

Vlad Ayzenberg @vayzenb.bsky.social · 22/07/2026

Excited to share our review in @cp-neuron.bsky.social with @lauriebayet.bsky.social and @mickbonner.bsky.social! We describe how implementing principles from child development can advance the mechanistic plausibility and capacities of AI models We packed A LOT into this review, here's a quick 🧵

17030

Reposted by Francesco Poli

Danny Kingsley @dannykay68.bsky.social · 20/07/2026

EXCELLENT: The Netherlands new OA policy seems to be well thought through obc.copim.pub/when-scaling... Policy explicitly recommends redirection & reallocation of funds from existing ‘big deal’ budgets towards community-driven alternatives, incl university-led diamond OA & Netherlands Uni Presses

obc.copim.pub

When 'scaling small' becomes national policy

The Netherlands' new open access strategy echoes the principles the Open Book Collective was built on - 'scaling small' and community-led publishing

17333

Reposted by Francesco Poli

Ben Hayden @benhayden.bsky.social · 20/07/2026

In Tye et al., 2024, one of the authors (I assume it was Fusi) says they were accused of ‘‘'turning cortex into a bowl of porridge'’ at a major meeting". To me, that's what a paradigm shifting sounds like. www.cell.com/neuron/fullt...

cell.com

Mixed selectivity: Cellular computations for complexity

Neurons capable of mixed selectivity enhance the brain’s computational capabilities by responding to various inputs in complex ways, beyond simple combinations. Here, we describe possible neural mecha...

0136

Francesco Poli @francescopoli.bsky.social · 14/07/2026

In the furure, I hope to see as many sessions focused on theories, or on methods, as there were sessions focused on data this year. Data is amazing, and naturalistic data is the ultimate test of our theories (when we will get there), but the imbalance was stark.

030

Francesco Poli @francescopoli.bsky.social · 14/07/2026

Of course, it might be me, and I might shortshighted here. But for now, what I really appreciated was strong experimental work set out to test precise theories of development and cognition. There was plenty of this too, and I particularly liked the talk by Ghislaine Dehaene-Lambertz in this regard.

120

Francesco Poli @francescopoli.bsky.social · 14/07/2026

For a deeper understanding of cognitive development, we need better theories and better methods. I also think we need old-school experimental work in the lab, because it's the only way to cleanly test competing hypotheses about development and disentangle variables that are counfounded in real life.

130

Francesco Poli @francescopoli.bsky.social · 14/07/2026

I worry that a field so focused on collecting data will become a data-driven field, rather than a theory-driven field. We describe phenomena ("Look, toddlers in Hong Kong learn to pour tea!"), but we do not reach any deeper understanding of them.

130

Francesco Poli @francescopoli.bsky.social · 14/07/2026

Some talks featured videos of babies in natural environments followed by ZERO statistical analyses. Presenters might have skipped them for lack of time, but a deeper reason for their absence is that many studies were descriptive, rather than aimed at understading or explaining cognitive development.

140

Francesco Poli @francescopoli.bsky.social · 14/07/2026

There is no way around this: naturalistic data is exteremly complex data, and so it requires extremely complex analyses. Of course there were a few examples of really good computational approaches to handle these data, but they were the exception rather than the norm.

131

Francesco Poli @francescopoli.bsky.social · 14/07/2026

The incredible step-up in both quantity and quality of real-life data collection (head-mounted eye-tracking and cameras, neuroimaging in home environments, etc.) is not matched by a step-up in terms of the computational analyses those data actually require.

130

Francesco Poli @francescopoli.bsky.social · 14/07/2026

It's quite clear that the field is moving more and more towards research in naturalistic, ecological settings. Overall, I think this is great: we eventually have to predict how infants think and behave in the real world. However, I don't think we are ready for that yet. And the main reason is that..

140

Francesco Poli @francescopoli.bsky.social · 14/07/2026

I went to ICIS in Panama last week, and the long trip back gave me time to think about how our field is doing and where it's heading. My general take is: there's not enough theoretical and methodological work in our field - but there's also reasons to be hopeful. A thread 🧵

2172

Francesco Poli @francescopoli.bsky.social · 13/07/2026

New Preprint! We gave neural networks infant-like curiosity and let them free to explore. They grew brain-like architecture, followed human synaptic development, and even started to display compositional thought. So we think curiosity doesn't just come from a complex mind... it helps us build one 👇

biorxiv.org

Curiosity shapes brain-like architectures and functions

How does complex cognition emerge from simpler underlying processes? We show that two components are sufficient: infant-like curiosity and brain-like biophysical constraints jointly drive the emergenc...

17628

Reposted by Francesco Poli

Jessica Sommerville @jessicas.bsky.social · 07/07/2026

#ICIS2026: Terrific symposium this morning on the motivations underlying persistence in infants and toddlers with @gracewujiameisun.bsky.social @jaeminhwang.bsky.social @francescopoli.bsky.social @amyneedham.bsky.social

0102

Francesco Poli @francescopoli.bsky.social · 24/06/2026

All the students are using this tech and they don't know how to use it though :( failing to adapt because the tech might disappear in the future is as foolish as tech companies investing in it for what it might buy them in the future. In the meanwhile the present is f'd up.

030

Reposted by Francesco Poli

Dr Charlie Gardner @charliejgardner.bsky.social · 23/06/2026

Reform UK councillors have introduced a motion to rescind King's Lynn and West Norfolk's climate emergency declaration and deprioritise decarbonisation efforts The council were going to vote on it on Thursday But the meeting has been postponed due to extreme heat

451231642

Francesco Poli @francescopoli.bsky.social · 13/06/2026

Why do so many interventions target parents or children directly, enforcing the belief that they did something wrong? Is wealth redistribution the only valid intervention?

073

Reposted by Francesco Poli

Michael Okun @michael-okun.bsky.social · 10/06/2026

If human editors can’t control who reviews science, it’s no longer peer review — it’s a rubber-stamp machine designed for volume and profit, not quality. I have no intention of attaching my name to it. So I’m out.

302361386

Reposted by Francesco Poli

Jonathan @jmcphers.bsky.social · 10/06/2026

When we started working on Positron, we began by asking "what did we always want to do in RStudio, but couldn't"? Years of community upvotes on RStudio's issue tracker helped us form that wishlist. Here's how some big ideas made their way to Positron: opensource.posit.co/blog/2026-06...

opensource.posit.co

RStudio's Top Feature Requests ... In Positron

Many of the most upvoted RStudio feature requests are already implemented in Positron, and we explore ten of them.

33911

Reposted by Francesco Poli

Francisco Rodriguez-Sanchez @frodsan.bsky.social · 17/05/2026

#StatsPubs No matter how many times you read it, this paper on hierarchical GAMs keeps giving. Clear model explanations together with R code (mgcv). So useful! doi.org/10.7717/peer... Thanks for writing @ericjpedersen.bsky.social, D. Miller, @gsimpson.bsky.social @noamross.net

In this paper, we discuss an extension to two popular approaches to modeling complex structures in ecological data: the generalized additive model (GAM) and the hierarchical model (HGLM). The hierarchical GAM (HGAM), allows modeling of nonlinear functional relationships between covariates and outcomes where the shape of the function itself varies between different grouping levels. We describe the theoretical connection between HGAMs, HGLMs, and GAMs, explain how to model different assumptions about the degree of intergroup variability in functional response, and show how HGAMs can be readily fitted using existing GAM software, the mgcv package in R. We also discuss computational and statistical issues with fitting these models, and demonstrate how to fit HGAMs on example data. All code and data used to generate this paper are available at: github.com/eric-pedersen/mixed-effect-gams.

216252

Reposted by Francesco Poli

Ricard Solé @ricardsole.bsky.social · 25/11/2025

Why do brains generalize so well while today’s neural networks often fail outside their training data? Check this paper in @cp-neuron.bsky.social argues how neuroscience can guide better architectures & representations. Check the table below. www.sciencedirect.com/science/arti...

2357

Francesco Poli @francescopoli.bsky.social · 30/04/2026

If you want to keep up to date with super cool developmental research, give a follow to @milestoneslab.bsky.social!

021

Reposted by Francesco Poli

Yaniv Abir @yanivabir.bsky.social · 18/04/2025

What drives human curiosity? Is it a need to balance stimulation — or something we learn over time? In our 🚨 new preprint, we show that learning reinforces curiosity, especially for related content. osf.io/9bw6j_v2 w/ Jane Mok, @chrisbaldassano.bsky.social , Caroline Marvin, Daphna Shohamy 🧵👇

osf.io

OSF

1228

Reposted by Francesco Poli

Daniel 🕹️ @strengejacke.de · 26/04/2026

🚀 New vignette: Uncover "Context Effects" in mixed models. Stop blending within- & between-person effects. Learn to demean variables and formally test state vs. trait differences side-by-side with the #RStats #easystats {modelbased}📦 easystats.github.io/modelbased/a...

easystats.github.io

Context Effects: How the Environment Shapes the Individual

14812