iiinova — where your digital self keeps evolving

Reposted by Julian Berger

Philipp Lorenz-Spreen @lorenzspreen.eurosky.social · 15/10/2026

Join us in Dresden in May for the @netsciconf.bsky.social we are working hard on it, and promise it will be a blast! Call for abstracts and satellites out now, submit your work here 👇

083

Reposted by Julian Berger

Max Planck Institute for Human Development @mpib-berlin.bsky.social · 15/10/2026

Human–AI teams may be more effective than previously thought. 🤖🤝 A study led by researchers @arc-mpib.bsky.social finds that learning is key: when people receive both feedback and AI explanations, human–AI teams can outperform the stronger partner alone. 👉 www.mpib-berlin.mpg.de/press-releas...

0107

Reposted by Julian Berger

mark brandt @mjbsp.bsky.social · 02/10/2026

How do people make decisions in complex conjoint experiments of policy choices? When decomposing individual responses into different strategies, most people use a heuristic strategy. Personally, I was surprised how many people are using the rational strategy! www.pnas.org/doi/10.1073/...

02612

Reposted by Julian Berger

Copenhagen Center for Social Data Science (SODAS) @cphsodas.bsky.social · 28/09/2026

Can human-AI teams outperform either one by themselves? @officialberger.bsky.social, @jasonburton.bsky.social and co. found that they can, but it's important that the human gets explanations for the AI's answers and a chance to learn from wrong decisions. More in PNAS: www.pnas.org/doi/full/10....

pnas.org

PNAS

Proceedings of the National Academy of Sciences (PNAS), a peer reviewed journal of the National Academy of Sciences (NAS) - an authoritative source of high-impact, original research that broadly spans...

022

Reposted by Julian Berger

Florian Keppeler @flokeppeler.bsky.social · 27/09/2026

💡 Currently, it looks like we judge human–AI teamwork before anyone has had a chance to learn it. This is a key direction for future research. 🔬 Wilma, PhD (Pretty helpful Dog) and #postdog 🐾 @au.dk 🔗 doi.org/10.1073/pnas... End🧵

doi.org

Human learning is an understudied but promising lever for boosting human–AI synergy | PNAS

Humans collaborating with AI hold the promise of achieving superior outcomes compared to either acting alone (i.e., human–AI synergy). However, the...

033

Reposted by Julian Berger

Valerii Chirkov @chirkovv.bsky.social · 24/09/2026

New paper alert! 🚨 How does social information quality impact collective intelligence in human groups? We used immersive 3D experiments to find out! Out now in Nature Comms 🧵👇 Read it here: www.nature.com/articles/s41...

nature.com

Payoff selectivity drives collective intelligence during dynamic resource tracking in humans - Nature Communications

Humans rely on complex social learning and collective adaptation to navigate dynamically changing environments. The authors show that human collective intelligence in a dynamic, naturalistic task depe...

2187

Julian Berger @officialberger.bsky.social · 22/09/2026

Setting humans up for success is indeed what we need. It requires a more systematic study to understand what drives synergy — we should include more diverse study designs that actually benefit human cognition

030

Julian Berger @officialberger.bsky.social · 22/09/2026

New paper out in @pnas.org 🚨 Human learning is an understudied but promising lever for boosting human–AI synergy @jasonburton.bsky.social @ralfkurvers.bsky.social @stefanherzog.bsky.social @dirkwulff.bsky.social

22213

Reposted by Julian Berger

Julia M. Rohrer @dingdingpeng.the100.ci · 18/09/2026

This is an *excellent* post on cluster analysis. If you ever thought about applying cluster, latent class or latent profile analysis to psychological scales, you absolutely should read this!>

512335

Reposted by Julian Berger

Berna Devezer @devezer.bsky.social · 14/09/2026

This is what I've been saying for years. That's also why I'm no more than lukewarm towards multiverse analyses on experimental data. If you're going to spend all that computational power, why not do it a priori so you can design the best experiment under well-articulated assumptions?

3376

Reposted by Julian Berger

Ed Hagen @edhagen.net · 11/09/2026

Well, it was free for one thing 🙂 Maybe @officialberger.bsky.social can comment on the value-added beyond Claude.

112

Julian Berger @officialberger.bsky.social · 11/09/2026

So you can spend all your Fable/Astra on this but we’re testing open-weight models rn and cost have reduced by 90%. This also ensures zero retention, processing in EU and GDPR compliance. Could be good for yet unreleased results. Hope to make that accessible soon, working on it as of writing this

000

Julian Berger @officialberger.bsky.social · 11/09/2026

In general, I’d like everyone to use such harnesses like Code/Codex (responsibly). But maybe there is a more convenient and more accurate way than running this locally and fiddling with your own prompts all the time. Also, by no means do we need the big costly models, at least for reproducibility.

100

Julian Berger @officialberger.bsky.social · 11/09/2026

Cheers 👋 We’re building rigor to (i) establish a baseline of how well LLMs can reproduce research without errors, (ii) reproduce papers en masse and (iii) offer automated reproducibility assessments for peer review. It’s really about benchmarking and then offering a one-button solution, IMO.

101

Reposted by Julian Berger

Jamie Cummins @jamiecummins.bsky.social · 10/09/2026

Machine learning and generative LLMs are subject to many of the same drawbacks and issues as general empirical research in the typical publishing system. In this piece now published at NHB we argue that, like for empirical research, ML and LLMs can benefit from the use of Registered Reports.

02210

Julian Berger @officialberger.bsky.social · 08/09/2026

Thank you, @johannesbreuer.com. We’re very busy behind the scenes right now but a big new release of rigor is coming soon ❤️

150

Reposted by Julian Berger

Andrew Gelman et al. @statmodeling.bsky.social · 02/09/2026

Metascience corner: What can we learn from 12 million empirical research results? statmodeling.stat.columbia.edu/2026/09/02/m...

statmodeling.stat.columbia.edu

Metascience corner: What can we learn from 12 million empirical research results? | Statistical Modeling, Causal Inference, and Social Science

083

Reposted by Julian Berger

Levin Brinkmann @levinbrinkmann.bsky.social · 18/08/2026

We worry about homogenisation, but can AI also expand human culture? In a multi-generational experiment we demonstrate AI-induced cultural shifts. Machines lastingly shift human behavior, beyond their presence, just as AlphaGo did to Go. www.nature.com/articles/s41...

nature.com

Propagation and preservation of AI-discovered problem-solving strategies in human culture - Nature Communications

AI can uncover problem-solving strategies beyond human discovery. Here, the authors show that AI-discovered strategies propagate and persist in human populations, producing cultural shifts when non-tr...

12213

Reposted by Julian Berger

Ralf Kurvers @ralfkurvers.bsky.social · 13/08/2026

Proud of the whole team, esp of @alexschakowski.bsky.social for putting together this review: Toward a psychology of real-life decision-making. www.sciencedirect.com/science/arti... @mpib-berlin.bsky.social @scioi.bsky.social @arc-mpib.bsky.social

sciencedirect.com

13019

Reposted by Julian Berger

Dirk Wulff @dirkwulff.bsky.social · 03/08/2026

🚨 Out now in TICS How can language models help cognitive science? @ruimata.bsky.social & I outline 5 uses: mapping research fields, formalizing theories, cleaning up constructs & measures, predicting behavior across tasks, and capturing environmental variation. 🔗 doi.org/10.1016/j.ti...

03812

Julian Berger @officialberger.bsky.social · 04/08/2026

Cool work! You might be interested in checking out rigor.me

rigor.me

Rigor

121

Julian Berger @officialberger.bsky.social · 23/07/2026

Happy to support reproducibility efforts whenever we can. Reach out if you want to use rigor.me at scale or in teaching!

130

Reposted by Julian Berger

Katelyn Nutley @katelynnutley.bsky.social · 23/07/2026

Reminder: tomorrow's the Reproducibility Hackathon @edfuturesinstitute.bsky.social! 💻🎉 Big thanks to our pals @mpib-berlin.bsky.social for letting us beta-test their new tool, rigor.me. We'll be reproducing a paper and testing our own code's reproducibility. Open science, one step at a time 🔬

153

Reposted by Julian Berger

Martin Modrák @modrakm.bsky.social · 17/07/2026

Multiverse analysis, abdication of responsibility and manufacturing of doubt: I have written on some downsides I see with multiverse analysis (which I like in principle): arxiv.org/abs/2607.14623 I was inspired/provoked to write it by @dingdingpeng.the100.ci (thanks!)

Abstract: I argue that multiverse analysis is highly suited to two undesirable uses: abdication of researcher's responsibility for their conclusion and manufacturing of doubt. A review of multiverse analyses published in 2025 provides tentative empirical support that abdication of responsibility is present in the literature and I mention anecdotal evidence that multiverse has been used for manufacturing of doubt about Covid-19 precautions. To mitigate negative effects if multiverse analysis becomes widely used I suggest the community adopts two conventions for evaluating multiverse analyzes: evaluating multiverses by the single worst universe they contain and considering large size of a multiverse as a sign of weakness rather than a praiseworthy achievement.

45717

Reposted by Julian Berger

Jamie Cummins @jamiecummins.bsky.social · 14/07/2026

Today we launch the first stable release of RegCheck: v1.0.0. RegCheck makes it easier and quicker to compare study registrations to published papers for consistency - something we know is important in principle, but rarely done in practice. A 🧵 on what's new: regcheck.app

regcheck.app

RegCheck

RegCheck is an AI tool to compare preregistrations with papers instantly.

512883

Reposted by Julian Berger

Sandra Grinschgl, (she/her) @grinschglsandra.bsky.social · 09/07/2026

Research on cognitive offloading × AI is booming and getting lots of public attention. Unfortunately, some (or even many) of it raises serious credibility concerns. Like this paper, which others and I recently commented on via @pubpeer.com: pubpeer.com/publications...

pubpeer.com

PubPeer - AI Tools in Society: Impacts on Cognitive Offloading and the...

There are comments on PubPeer for publication: AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking (2025)

1178

Reposted by Julian Berger

Joe Bak-Coleman @jbakcoleman.bsky.social · 08/07/2026

Every once in a while you find a paper that does a lot with a simple approach. Today's is from @natematias.bsky.social, Cassidy Waldrip, and @davidlazer.bsky.social. They simply ask how much computational social science can be reproduced? #metascience. journalqd.org/article/view...

journalqd.org

View of Commercial Determinants of Replication Infeasibility in Computational Social Science

2258

Reposted by Julian Berger

Eiko Fried @eikofried.bsky.social · 06/07/2026

Come work with me, Janna, Jonas, and Bart at Leiden University :). #MentalHealth #Psychology #PsychSciSky #academicsky

02915

Julian Berger @officialberger.bsky.social · 02/07/2026

We have some slots available tomorrow and over the weekend if folks want to test how reproducible their work is! Get a free reproducibility check of your paper: rigor.me I would especially welcome some neuro & ML & econ folks to test this. Throw any weird data at it. Much appreciated ❤️

rigor.me

Rigor

000

Reposted by Julian Berger

Dirk Wulff @dirkwulff.bsky.social · 01/07/2026

🚨 Now out in PNAS In decision research, "talk is cheap." We show it isn't. Using LLMs to analyze participants' free-text explanations of their choices, we find verbal reports are a rich, scalable window into how people actually decide. 🔗 www.pnas.org/doi/10.1073/... Led by Kamil Fulawka

04215

Julian Berger @officialberger.bsky.social · 01/07/2026

I am tentatively excited about Claude Science if it enables more robust coding and reporting going forward, but also here, that was doable before with Code. It also invites to walk all paths in the garden of forking paths and explore the multiverse until one finds a story that one likes.

100

Julian Berger @officialberger.bsky.social · 01/07/2026

Yep, saw and tried that yesterday. Same results as us. Maybe we are on to something with Rigor, eh? COS, I4R and others are working on similar projects. For checking results upon submission we may need more, though, than individuals playing with Claude locally. That was doable with Cowork/Code, too.

130

Julian Berger @officialberger.bsky.social · 30/06/2026

Well that’s perfect! Thanks for pouring it out, we’ll get it sorted

010

Julian Berger @officialberger.bsky.social · 30/06/2026

All of the code, unlikely. Prompts, maybe. We're working on making this as accessible and transparent as possible to as many folks as possible but we are definitely incurring costs to run this. Would love to hear some ideas about this, happy to chat!

100

Julian Berger @officialberger.bsky.social · 30/06/2026

Fable is not even needed, I think. Opus is giving us the best results rn but there are ways to slim it down and rely on smaller models in parts, thereby reducing costs. We're working on that, first we built it to not miss anything and not make errors. So, if you find some, please let me know!

200

Julian Berger @officialberger.bsky.social · 30/06/2026

Yes and no. Its a series of hardwired steps, especially for setting up a docker compute environment. Then we do have a series of LLM calls for execution, indeed. One-shotting, so far, was missing a lot of information and context and created variability between runs.

100

Reposted by Julian Berger

Ed Hagen @edhagen.net · 30/06/2026

Damn! I ran my just-published paper through it and the review was awesome, if a bit humbling. I agree with most of it, and I would have addressed several of the critiques if I had this before submitting. Great stuff!

131

Julian Berger @officialberger.bsky.social · 30/06/2026

Awesome, thanks for giving it a shot! I think the AI review is very harsh. But by being harsh, it can be helpful? Want to try another paper? Anything that did not work for you?

110

Reposted by Julian Berger

Matti Vuorre @matti.vuorre.com · 29/06/2026

Usability 10/10 (though of course more speed = more better but I understand this takes time and compute isn't infinite). I will have to carefully check the discrepancies but I immediately understood some of them. Fortunately there was nothing major this time :)

211

Julian Berger @officialberger.bsky.social · 29/06/2026

Please let me know if you find any errors. Especially false positives where we say some results of yours do not reproduce but they do on your computer.

000

Julian Berger @officialberger.bsky.social · 29/06/2026

Yes, so, for the beta Philip and I do technically have insight into your results to make sure we can fix any errors on our end (e.g., something did not run). You can share your results with other folks via a link. But its not public! Also, we will delete all data after the beta (~4 weeks).

000

Reposted by Julian Berger

Matti Vuorre @matti.vuorre.com · 29/06/2026

I tried this on one of our papers from 2021. It did a fantastic job! For example (see screenshot) it verified our reported sample size & response rate, and about 80 other reported numbers/statistics with descriptions of where, why, and how some numbers differed.

151

Julian Berger @officialberger.bsky.social · 29/06/2026

Awesome, thanks for trying it out!! Would love to hear some feedback wrt how usable you think it is and what you think is missing. Let me know if you want to try more papers :)

100

Reposted by Julian Berger

James Feigenbaum @jamesfeigenbaum.bsky.social · 29/06/2026

Like the wallet inspector, but for your data and code. (Kidding, this is very cool and useful)

052

Julian Berger @officialberger.bsky.social · 29/06/2026

I should also say that we have found errors in published work and the original authors confirmed misreporting / typos.

110