the Google AI summaries are ... remarkable. to paraphrase Douglas Adams, they are almost but not quite completely unlike useful information

Perhaps, given how much hype and skepticism there still are around LLMs, we do not have enough papers describing embeddings from first principles. Honestly, I can go either way here. Either wr have too many or too few papers

Does anyone in #StatsSky or #EconSky or whatever know of anyone has published about an analogue of the competing risks identification problem in space rather than time?

🤦‍♂️ 10/10

This headline is causing my eye to twitch. Causal inference nerds are going to have a field day

Well now

I think MOOCs are a bit underrated here, but this likely has to do with me and Prof Dynarski considering different outcomes. I can go on YouTube ( or EdX or whatever) and take entire courses FOR FREE. For a certain type of student (definitely a minority of them) that is a fantastic resource

Not the point, but it might behoove them to think through why the Catholic Church has "revenue in the hundreds of billions, perhaps even trillions, per year". What do they think the Catholic Church does with that money?

22 Short Films About Springfield first aired on 14th April 1996, making Steamed Hams thirty years old today.

Day 2345 of the discourse. The combat posters have reached a stalemate, neither side advancing nor retreating. Trenches are being dug, foxholes prepared. The war will continue.

Still one of my favorites:

Someone compiled all of the Calvin and Hobbes strips about Calvin’s polls of his Dad, which is proof that good people do exist on the internet imgur.com/a/poll-resul...

Finally finished up a blog post on estimating Bradley–Terry models using brms. www.m-flynn.com/posts/2025-1...

Who the best nerdy wimp action hero?

Me, watching a trailer for The Furious: "Why do they always kidnap the family of the guy who's built like a tank and spends a lot of time in the gym? Where are the pot-bellied dads with skinny arms?" Anyway looks fun (warning: violence): youtu.be/Y35swNcw6gw?...

Neat! I will need to look more closely at this

Are joint predictive models enough, or do you actually need casual inference? notstatschat.rbind.io/2026/04/13/a...

Kid just asked who won WWII and if the Nazis were bad guys. A lifetime of dad books has prepared me for this moment...

When I woke up this morning I did not think I would be posting thirst traps of Ryan Reynolds, yet here we are

Interesting (and fairly good) article on the manosphere. Now, a good question for discussion: Is the manosphere a market failure?

Help Mircea

On the discourse around "AI" I really stand by this the more I learn - transformer models have a really cool "encode/decode" structure that makes for a great model of language (particularly if that model is large) but is very different from how we map map words onto our "world models" when >

We know models aren’t great at counting the “r”s in “strawberry” but why? The answer isn’t as simple as tokenizers!

Sometimes you walk by a statistics class and the professor is writing like "y = a + bx + e" on the board and everyone is taking notes like its an actual school

It's interesting to see how risk tolerance and acceptable mitigation change over time.

I disagree. Like everything else it will take some time to adjust, but profs will figure out how to assign and grade long form writing.

They have voice! Comms are back and trajectory is good. 🎉🥳🙌

My tongue-in-cheek solution is to make manly jobs more "girly" by lower their wage

There is no obviously correct benchmark for what that replication rate should be (and be skeptical of anyone who is sure they know what it is). Heck, with all the time I've spent in this work, I am still truly not sure if/how much we should be bothered by the rates found.

👀

Hey, survey nerds!

Interesting distribution for the results

Best Star Trek crew, wrong answers only:

Correct. In addition, ML/AI works best when, at a minimum: 1) your outcome variable is clearly defined and cleanly measured 2) your input variables are clearly defined and cleanly measured 3) both are stable over time and environment, and there are few edge cases

1. slop is a normal good, maybe even more as the population interacting with it tilts bot 2. "I don't care who this offends, but" feels similar in meaning to "I'm sorry, but". kinda interesting how popular that mode is in boomerslop

😬

The Simpsons is the gift that keeps on giving

Ravens are cool too, for that matter www.cbc.ca/news/canada/...

Parrots are cool and "stochastic parrots" should be laudatory

Imagine being a historian and getting mad that Greenland Sharks have LinkedFin, so you sign up for LinkedZinn

An interesting read to which I both agree and disagree (eg, I do not see survey weighting as simulation). As always, give it a read. 1/N

Just saw this new paper by Brauer on how linear models of mean effect can be very misleading for ordinal data link.springer.com/article/10.1...

New pre-print! I cover a range of open capture-recapture models (single survey/robust design, multistate/multievent, in (Cormack-)Jolly-Seber variants) in Stan, and provide efficient log likelihood functions. I also introduce a method to account for unequal survey intervals in the entry process.

Stats nerds, assemble!

Time for this again

wut

Now is your chance, Canadians! Go tell Transport Canada why modern headlights suck www.cbc.ca/news/canada/...