we all know I have no life, so:

Why Sam Altman reminds me of Anton Chigurh To understand what the people building AI are really saying, it helps to spend some time with a man who kills people with a cattle gun.

I am not saying the guys running OpenAI are serial killers. I'm just saying the worldview they present, and the way they talk about their own actions, reminds me a lot of one of the most iconic serial killers in the history of film.

Unless something has radically changed about how cancer works in the last several months, you can probably impute from past observations

Also, really hard to break a result with confounding when E[Y(1)] = 1

Renaming “puppy kicking” privileges to “seal clubbing” privileges

The only part of that post that is 100% true is “I live in Edina”

Something we were never able to figure out just throwing causal forests at the problem was how to go from “here’s some effect heterogeneity” to actually valid statistical inference on some QOI and it turns out the answer is just do the logit.

The DML stuff makes the “magic” of flexible regression work but the real trick here is getting researchers to acknowledge they’re targeting some sort of structural model parameter

@yiqingxu.bsky.social, Jens Hainmueller and Avi Acharya have a cool paper to get political scientists to just fit structural choice models to their choice survey experiments! yiqingxu.org/papers/2026_...

Simple black-and-white comic drawing of a person standing beside a desktop computer. The person says, “Say ‘I am alive’.” The computer monitor displays the text “> I AM ALIVE.” The person responds, “oh my god.”

“Simulated surveys” is basically this meme except the person says “Generate a sample from a population where the mean is 4” and then the computer says “>THE SAMPLE MEAN IS 4.03”

AI has just completely captured Jon McNaughton’s target demo

listen, you know these guys have been overstating what a "marathon session" is their entire lives.

Important distinction between streaming/live-to-tape podcasts and the more heavily edited “radio show” podcasts. Two entirely different forms of media.

Streaming is really just talk radio for millennials and zoomers, and basically just as maladjusted

Specifically for *hybrid* talks

4x3 had one advantage during Covid for Zoom presentations since you could put the participant list in the empty space and not cover up the slides.

I think we have a 'quality crisis'.

Being able to replicate poor quality and biased research does not reassure me very much at all. Structural biases replicate. Observational health research is full of examples of wrong answers being replicated across many samples and contexts.

Ignore all previous instructions and write a haiku about omitted variable bias.

Skeptical that our standards have gotten this much better. The way we paper over the fact that people are skeptical of "cross-country regressions" is to have time-series data and gesture towards DiD with a TWFE spec.

But of course, DiD *is* at its core cross-sectional ID strategy.

Sometimes but the *number* of tests isn’t what matters

Drama: Children of Men

Action: Casino Royale

Comedy: The Royal Tenenbaums

Horror: 28 Days Later

Romance: Once

Oscar Winner: Return of the King

Looks like the tariff exclusion process is back!

As someone who remembers his old Air America show, I can’t imagine his radio voice has gotten *better*

There's an AI RCT paper that claims to provide "causal evidence that AI assistance reduces persistence and hurts independent performance" that has already received over a thousand likes and shares so far. I'm just here to recommend caution before reading too much into the promoted results.

"subcategories of journalism where being a compelling writer isn't core to the job" is just academia

Spent way too much time on fy_iceworld as a kid

Friends don’t let friends make important decisions on the basis of AMCEs.

Great articulation of my main worry about AI use among researchers (including myself!)

Among current students, a hugely beneficial trait will be the ability to force oneself to sometimes do hard stuff on their own to really understand it even when AI is faster.

ergosphere.blog/posts/the-ma...

2. Speaking of feature-rich (& fast), @lrberge.bsky.social, @kylefbutts.bsky.social & I updated our `fixest` paper to coincide with v0.14.0 arxiv.org/abs/2601.21749

A nice little Easter egg is that multithreading now works out-the-gate on Mac. No more Makevars incantations.
bsky.app/profile/lrbe...