One of my favorites paper got published 🤓 It covers a lot of ground and it’s the best summary of my views on misinformation and what to do about it. Give it a read :)
🔓 osf.io/preprints/ps...
👉 doi.org/10.1177/1461...
Only heard the french version 😄
New preprint out today (osf.io/preprints/ps...). We tested whether AI agents are actually infiltrating online surveys.
Spoiler alert: they aren't
Thread 🧵
[1/9]
Thanks Tanay 😊
Merci Jérémie 😊
More in the paper :) Happy to chat about it!
Paper (free access for 50 days): authors.elsevier.com/a/1mqZS2Hx32...
Data, code, materials, and preregistrations are open: osf.io/aecyr/overview
The same Bayesian framework explains information search. Participants flexibly queried new information in ways that match an optimal information-theoretic search model.
A Bayesian model best explains these inferences. Participants’ judgments are best captured by a model that rationally integrates new evidence with prior expectations about others’ competence.
Key findings:
1. People can accurately predict what others know from very limited information. From a single observed answer, participants could predict the probability that someone would know another trivia question with striking accuracy!
After seeing an individual’s performance on a trivia question, participants had to either:
(a) predict the individual’s ability to answer other trivia questions from the same theme, or
(b) select which information would be most diagnostic for inferring an individual’s competence
In this paper, we examine how people infer others’ competence from sparse evidence, and the computational principles underlying these judgments.
We tested participants’ inferences on two tasks requiring fine-grained estimations.
If someone knows that Venus is the only planet in the Solar System that rotates clockwise, will they also know what Earth’s only natural satellite is? What about which planets have no moons at all?
It turns out people are remarkably good at figuring this out.
Thrilled to share that our new paper is now out in @cognitionjournal.bsky.social: "Who knows what? Bayesian Competence Inference guides Knowledge Attribution and Information Search," with @oliviermorin.bsky.social , @hugoreasoning.bsky.social & @tadegquillien.bsky.social!
Link: tinyurl.com/ykyhxcc6
Thanks for sharing 😊
Thanks for the thread, I was suspicious seeing the headlines but hadn’t got the time to dig in.
Also, the posts I saw about the report use a strong causal wording for the relationship between social media and wellbeing, is it the same for the WHR or are they more cautious?
The World Happiness Report 2026 theme is social media and adolescent wellbeing, so I tried to have a read. It feels a mixed bag to me.
They gave prominence to the view of Jonathan Haidt and Zach Rausch, which tells about the framing, but there are also clearly more nuanced chapters.
In poverty, do people take more or less risk? Some theories contend that they avoid risk out of caution. Others that they take risks (e.g. crime) out of desperation.
In our new paper in BBS, we show that they are the two sides of the same coin: the desperation threshold.
Peer commentary call soon!
CALL: a PhD grant (3 years) to do a PhD with me at @cognitionens.bsky.social on the evolution of graphic codes. euraxess.ec.europa.eu/jobs/410213
🎉 Our paper is out in Communications Psychology: a perspective on behavioural public policies & the psychology of poverty.
Core idea: poverty can shift psychology, so the same intervention won’t work the same for everyone.
Thread 🧵
Or simply good old Rmarkdown
A digital illustration styled to look like an embroidered fabric patch or cross-stitch. The text reads 'TWO YEARS OF WORD GAMES' in the center, with '2025' at the bottom and 'AMAN V. RHEA' in the top right. The background is a checkered pattern of colored squares (green, yellow, gray, purple, and blue), mimicking the color palettes of Wordle and Connections.
A scatter plot titled 'CROSSWORD TIMES' comparing daily Mini Crossword completion times over two years. The chart uses a pixelated aesthetic. Aman's data (left, in blue) shows a widely scattered distribution with a trend line hovering between 1 and 1.5 minutes, and includes a high outlier labeled 'Legendary bad Saturday.' Rhea's data (right, in red) is much more tightly clustered near the bottom, with a trend line consistently sitting around or below the 1-minute mark.
A waffle chart titled 'WINS IN A DAY' displays the percentage of days when each player achieved 0, 1, 2, or 3 wins (a 'Sweep'). Aman's grid is dominated by '1 win' (yellow) and '0 wins' (beige), with only a few red squares representing a Sweep. Rhea's grid is much more colorful, featuring large blocks of '2 wins' (green) and a significant number of 'Sweep' (red) squares.
An infographic titled 'CONNECTIONS: Who flops, clutches, or cracks purple?'. A table shows Aman has a slightly higher perfect game rate (30%) than Rhea (29%). Below, a horizontal bar chart compares specific play styles: Aman is more likely to play 'By The Book' (solving in exact order, 13%) and 'Total Flop' (losing all attempts, 6%). Rhea is significantly more likely to 'Clutch' (win with 3 mistakes made, 17%) and solve 'Purple First' (cracking the hardest category first, 8% vs Aman's 3%)
Rhea and I have been tracking our @nytimes.com word games (Wordle, Connections and the Mini) for the past two years. I decided to analyze our 4,200 gameplays with #rstats to determine a winner and made some festively embroidered charts with ggplot. Have a read. #dataviz
aman.bh/blog/2025/tw...
Congrats Oleg !! 🎉🥳
Hi, many thanks for sharing this!
I've tested across several device (pc with low quality camera and a recent iphone). On the iphone it works well, but on the pc it works only on Firefox but it doesn't work on Brave (I tried several times). I don't understand why
It takes 225 million years or so for the Solar System to make one orbit around the galaxy. Sharks evolved long enough ago that they’ve done it twice.
@gregoryfiorio.bsky.social interesting !
Thanks Tanay !!
📄 Preprint: doi.org/10.31234/osf...
💾 Data, code, materials, and preregistrations: osf.io/aecyr/files?...
Huge thanks to my co-authors, all senior authors with equal contribution: @oliviermorin.bsky.social, @hugoreasoning.bsky.social, @tadegquillien.bsky.social
⚙️ We analyze behavior both at the aggregated level and at the trial level, showing that participants’ accuracy is not merely a wisdom-of-crowds artifact.
🤖 Our Bayesian model outperforms several plausible heuristic models and explains ~90% of the variance at the aggregated level.
More in the paper:
📊 We combine a rich behavioral dataset (N=1,916; 113,400 knowledge-attribution judgments; 2,960 search choices) with a large and diverse stimulus set (N >1200).
📌 The same Bayesian framework explains information search. Participants flexibly queried new information in ways that match an optimal information-theoretic search model.
