Razorbills have the slickest design in the animal kingdom. They look like they were made by Yves Saint Laurent

Good morning to you who've chosen not to use spaces and ampersands in your Windows username

The thing is, this doesn’t quite work. Data is about what you measure and how you interpret it. By basically removing the first thing (throw it all in) and the second (interpretation without context or expertise) AI doesn’t even necessarily look backwards.

lol when I moved from the UK to Canada I gave them the wildest exams. But my grade distribution was always perfect. Mostly because when you’ve got headroom in your instruments it’s easier to apply global weighting to make your assessments function more effectively.

Ok so Varla Jean Merman is phenomenal.

Going to the park to sit with my sisters and drink an iced tea

Up, up, down, down, left, right, left, right, start

RIP Magritte. You would’ve loved AI.

I believe this is the right general pattern: putting your name to an LLM output should expose you to the same professional, reputational, etc, consequences as if you had generated it in any other way. Any other pattern leads to a low-quality equilibrium. 1/ via @kirkpams.bsky.social

If I had one single piece of advice for people writing software today, it's "address your underinvestment in automated testing, even if you think you have good automated testing" It's old advice, but the leverage it gives you in 2026 is absurd.

Sad to see ticket sales for the World Cup are struggling. FIFA need to do something drastic to get people excited about it. Let the players pick up the ball

In "Uncertainty in Bayesian leave-one-out cross-validation based model comparison" doi.org/10.1214/25-B... we showed when LOO-CV elpd_diff and se_diff normal approximation uncertainty quantification in model comparison is well calibrated. We have now merged a related PR to loo R package 1/

Blue Film is far and away the best movie I’ve seen in a long time. Don’t look up anything beyond the very basic description because it unfolds into itself deliciously.

Nia DaCosta to Direct 'Sex Criminals' www.hollywoodreporter.com/tv/tv-news/n...

My issue is with amateur takes on information theory showing up in AI discussions. Basically, you have to know a lot more than the definition of entropy and its relevance to data compression.

I mean this is the CS/ML field in a nutshell. Conferences/arxiv have major downsides as tempo is traded for depth and correctness.

Act 2 of Cats.

the severe misread of youth by ai prophets is that at 22 you’re starving for agency. all your life you’ve waited to call your own shots. the idea that what you get instead is “tell the machine to draw a horse” is inherently offensive

A little ball of sunshine, a female Northern Yellow Warbler 🟡

I am dedicated to keeping the list that long. People liked my Twitter account and simply no. Not my joy.

Dogs make me want to get the no country for old men cattle gun. Little kids are cool.

It’s not the sort of work I ever did, but I have a lot of time for some of the thinking behind these model-free inference theories. (I do think they’re often over-sold and not allowed to live on their own merits in a desire to differentiate themselves. But the work is really helpful, model or no)

🚨A Researcher's Guide to Empirical Risk Minimization I put together a guide on regret theory for empirical risk minimization (ERM) as I understand it. The goal was to compile results and proof techniques I’ve found useful in my own work. I hope people find it useful more broadly

This is probably a better paper than the one I linked. But I’m wine-drunk in the afternoon arxiv.org/abs/1910.06386

I mean we understand it pretty deeply. This series of papers, for instance, gives more than you ever would want to know. Logistic regression (and other regular GLMs) work similarly. faculty.wharton.upenn.edu/wp-content/u...

Whoever decided that we as a derby community should use Discord as a primary communication platform a few years ago... I'd like to meet you in a dark alley. I just wanna talk.

It is not helpful to quote tweet someone talking about US funding cuts to point out that while his point may be substantively true, the graph he is illustrating it with lacks a constant denominator so it doesn’t show what he thinks. So I won’t do that. But my data sisters: denominator’s matter.

You know, I tried this on an arxiv paper a few weeks back. Just on a whim. It was the type of thing that would be at an AI conference. It made a great review. Unfortunately it missed that almost all of the key claims were full of technical errors. It was too busy being polite.

Haute & Freddy’s “Scantily Clad” is a banger AND I can’t help but think how it would be improved by Fred Schneider loudly shouting back the title after the hook.

Sometimes i forget that i have a very different base of cultural references to my colleagues. I needed to give notes after a presentation.

if your livelihood doesn’t depend on shipping code that works I am really not interested in your opinions about what LLMs are going to do to the software discipline tbh

The major reason why LLMs suck at writing C has been out in the open for years but nobody is ready to talk about it www.wired.com/story/openai...

It does this

NGL signalling is maybe the only use of sigma algebras. In times gone by, they often signalled that someone had done a measure-theoretic probability course but hadn’t understood it

Nothing will ever be funnier than this.

Use tools for things they useful for and not for other things should not be a radical take but somehow it is. You wouldn’t use a shovel to open a jar of marmalade.

Drag king name: Trey BN.

This is a hall of fame skeet. No notes. Using it in a future talk. I don't know what talk, but there will be one that uses it.

I whispered in his ear “I will murder you in your sleep”. He said “I love you”. It doesn’t change my plans.

I just got to tell my favorite Cristina Alberini story, where when she got told off for taking me and a seminar speaker from Yale to dinner at Jean-Georges, spat at the administrator "well we didn't have to spend any money on travel so I thought we could go somewhere nice for a change!"

Saw my friend @alexhanna.bsky.social (in sharpie drag because commit to every bit!) and her co-host Emily Bender (who was glorious) tape a Mystery AI Hype Theatre 3000 podcast. Main topic was www.vox.com/future-perfe.... (Second photo is 3/4 of a short dnd campaign I ran [alert: boyfriend on main])

do yourself a favor and search for something like "Pokémon" (something that was big in 2000 and is still big now). then do the same on Google It's amazing how much the Internet has changed - in many ways for the worse - and we don't even realize it

"I'll kick you in the balls until you forget your confirmation name"

But also... #statsmeme

blog post: work with us at Blue Rose ! use cutting edge statistics, machine learning, and engineering to study public opinion, forecast elections, and advise Democrats.

Truly one of the all-time great comedies. Italian Piñata will never not make me laugh.

New York!!! Mystery AI Hype Theater 3000 is joining you LIVE TONIGHT! At Starr Bar in Brooklyn! I'm literally on the M right now! Come through! www.eventbrite.com/e/mystery-ai... @emilymbender.bsky.social

Applied Statisticians, (violently)

2x sealed and a draft. Not pulling bad.