This doesn't work for confidence intervals that can be the whole real line. Ratio of Normal means (eg cost-effectiveness) is one genuine case (when both numerator and denominator have uncertain sign) and a trivial case is empty set with prob alpha and whole real line otherwise

Exactly this!

Tomas Hirst@tomashirstecon.bsky.social · 2w ago

It is not nothing that AI is quickly solving a lot of tractable but vast search space maths problems. It should tell you how useful it is when given a well defined thing to interrogate. That they are solved with simple prompts isn’t the point, it’s that the actual problem itself is well specified.

I've written a few books and checked all the quotes (deleting some bangers because I doubted their authenticity) and if you're letting fabricated ones slip through then you're cutting corners in an egregious way. It is not the copy-editor's job to check every source, it's yours. Do the damn work

Now that the entire tech industry is on board, here’s a call to build AI models not just with open weights but with an open corpus.

Public Intelligence

A call for AI models with open weights, open source, and open corpus.

argmin.net

Adding fuel to the fire, here's a post I wrote about open corpus models last July. It's all still true. www.argmin.net/p/an-open-mi...

An open mindset

The commitments required for fully open source machine learning

argmin.net

Ben Recht@beenwrekt.bsky.social · 2w ago

Now that the rich and powerful people are calling for open-weights models, it's time to pressure them to be legitimately brave and back open-corpus models.

Not sure about this. There is a middle class (in many countries), studying in higher ed not simply to survive. Of course there are socio-econ pressures, but there are also other socio-cultural tendencies that push people into opportunism at all costs.

Cat Hicks@grimalkina.bsky.social · 2w ago

Framing this as being about individual "caring" is basically the thing education equity has been pushing against for the last thirty years and I find it incredibly depressing how deficit mindset this platform is about learners

I'd even say *rarely* instead of not always. Also, not only psych and bio, I feel like many engineering disciplines do that too, albeit with a slightly different focus. Classic in engineering: use "Analysis 1" for thinning stud population, more applied parts for "real stuff". Bad bad bad.

Hendrik De Bie@debiehendrik.bsky.social · 2w ago

but we're also not always serving proper educational goals, or the field of math itself by doing so.

Let me close this thread by linking once more to the closest thing the mathematical community has to a collective position on AI and use of AI in mathematics, and the threats they pose: the Leiden Declaration leidendeclaration.ai •38/38

Leiden Declaration on Artificial Intelligence and Mathematics

This declaration calls for action to address the challenges posed by the use of artificial intelligence within mathematics research.

leidendeclaration.ai

That is very sad. I only knew him from some delightful email interactions and social media. And of course his books.

Dirk Eddelbuettel@eddelbuettel.com · 4w ago

Norm Matloff (1948-2026) @matloff.bsky.social I was made aware of the post below made by his daughter. We lost another friend and valued long-time contributor the #Rstats community, gone far too soon, and sure to be missed. His memory will be a blessing.

It feels a certain way knowing that @mjcrockett.bsky.social and I published the rebuttal to this article… six months ago.

AI Surrogates and illusions of generalizability in cognitive science

Recent advances in artificial intelligence (AI) have generated enthusiasm for using AI simulations of human research participants to generate new know…

sciencedirect.com

This skeet will self destruct@pardoguerra.bsky.social · 4w ago

This is quite the finding for a device thats just a plagiarism machine, or whatever we are calling it this week. www.nature.com/articles/s41...

The correlation coefficient is not a proper scoring rule. An agent who ranks models by correlation can be beaten by an infinite class of alternative rules. Here's a 2007 article on proper scoring rules: sites.stat.washington.edu/raftery/Rese...

This skeet will self destruct@pardoguerra.bsky.social · 4w ago

This is quite the finding for a device thats just a plagiarism machine, or whatever we are calling it this week. www.nature.com/articles/s41...

I hate this feeling

Olivia Guest · Ολίβια Γκεστ@olivia.science · 4w ago

It's so crazy... this as a situation erodes so much of our existing trust between each other I need to touch a lot of grass today as I feel so overwhelmed by the idea that anything I read or share could be generated — need some mental health reboot

Friends don’t ask friends to read and comment on what is evidently Claude’s work

Cannot stop laughing.

Julia M. Rohrer@dingdingpeng.the100.ci · last mo.

The children yearn for the mines.

Looking forward to reading it. Whenever anyone posts something sensible about multiple testing on bsky, you always see the Bayesians coming out of the underscrub...

Darren Dahly@statsepi.bsky.social · last mo.

This paper is worth it for table 1 alone. 👏 @sabinehoffmann11.bsky.social @boulesteixlaure.bsky.social et al onlinelibrary.wiley.com/doi/10.1002/...

A bit further in the article it basically says "doubly robust stuff + ipw GO BRRRRR" It makes me wonder whether Claude/gpt models would suggest that. Prolly. Do I want to know?

Saloni@scientificdiscovery.dev · last mo.

Oh no! Oh no!!

I wonder if people dismiss these warnings because they implicitly assume that if these vehicles were really *that* dangerous, then obviously government would regulate them appropriately.

Prof. Ian Walker@ianwalker.bsky.social · 2mo ago

🚨 New research by us! 📣 Most British drivers don't know how dangerous SUVs are to other road users, but adding information to SUV marketing materials seems to do almost nothing to change people's plans to buy SUVs Here's a 🧵 on why harder action is going to be needed 1/10

I wrote a thing about the simple case of the a.s. representation theorem, which is basically about random number generation notstatschat.rbind.io/2026/06/16/t...

"The increasing involvement of technology companies in mathematical research raises the risk that research questions may come to be prioritized because of their amenability to automated mathematics, rather than expert judgment of their deeper significance."

Leiden Declaration on Artificial Intelligence and Mathematics

This declaration calls for action to address the challenges posed by the use of artificial intelligence within mathematics research.

leidendeclaration.ai

Book: Test-Driven Data Analysis (by @njr0.bsky.social) Chapter 1 Data analysis as if the answers actually matter ht @cghlewis.bsky.social cc @rmcelreath.bsky.social book.tdda.info/book/chapter...

These are people who have never done anything like that in their lives. Their entire upbringing consisted of "nothing is ever your fault, it's always some nameless other".

Who needs Netflix when all the drama is on Bluesky...

Looking forward to this @royalstatsoc.bsky.social discussion meeting on July 1 featuring this paper on randomisation by @stephensenn.bsky.social. Paper: rss.org.uk/RSS/media/Fi... Please register: rss.org.uk/training-eve...

​Happy 20th Anniversary to #rdatatable! 🎂 ​Since 2006, DT[i, j, by] has been the gold standard for high-performance data manipulation in R. Speed, efficiency, and elegance! ​A huge thanks to Matt Dowle, Arun Srinivasan, and all the contributors! @rdatatable.bsky.social 🥂 ​#rstats #rdatatable

Nice and well-motivated exposition! One thing: I'd replace Wolfram something with Sympy or similar.

Clément Canonne@ccanonne.github.io · 3mo ago

Nothing incredible novel, but a short 🧵 thread about MGFs, their use, and why I love Taylor series. Say you want to compute the moments of a r.v. X, which is "sufficiently well-behaved" (in some rigorous sense). That is, 𝔼[Xⁿ], for some integer(s) n≥0 of your choosing. How do you proceed?

I was thinking exactly that when seeing those replies...

Olivia Guest · Ολίβια Γκεστ@olivia.science · 4mo ago

a lot of people reacting and saying other times technology was also forced on people as if that changes that the point is "this is very bad and maybe a worse case than previously"

They conducted a full test of its hacking prowess They evaluated its capabilities against current cyber defences And they published the whole evaluation in just over ONE week It is the only independent evaluation of this powerful new AI tool yet undertaken anywhere in the world 10/

Recording now available if you've missed it:

Natalia Ortega | Meta-analyses of dietary exposures must consider energy adjustment

YouTube video by the Causal Inference Interest Group (CIIG)

youtu.be

Georgia Tomova@georgiatomova.bsky.social · 5mo ago

Meta-analyses in nutrition may pool estimates for completely incomparable effects without realising it, just because of whether/how energy intake is adjusted. @nataliaortega.bsky.social will explain why, share results from a meta-scientific review & offer recommendations for ways forward 👇