I ♥️ that universities are scrambling over themselves to simultaneously end "DEI" while promoting "viewpoint diversity"

TBC, it's not particularly novel. And only one model has really impressed me. The local models are amazing, but their work is obviously worse... Good undergraduate paper-level

* here's a dataset

A whole section about possible exclusion restriction violations and a shift-share instrument. Robust SEs, fixed effects, robustness tests

I'm spending my Friday giving frontier local & API LLMs a really tough task:

"he's a dataset and DV, develop and test a theory and write it up"

No guidance or feedback

So far I've got a 2SLS with a plausible IV and AFAICT minimal hallucinated cites... Things are looking dicey, Team Humans

If you insist on reading Communion, make sure you read the one based on a true story

💯 the best essay about AI in academia yet, bar none

I mean, how hard could it be? ☢️

AI can enhance progress to our actual target, but it can also short circuit it in the same way that fraud and test cheating can, but at phenomenal scales

The problem isn't too much or too little AI in academia. It's that we've let pubs, a metric of our success, become the target. Our job isn't publish; it's create and communicate new knowledge

The irony of a DoD that talks about "re-establishing deterrence" all day long being deterred by a totally non-credible threat

New study finds notion that humans can "double-check" AI to be flawed. Participants trusted AI 93% of the time when it was correct, but still trusted it 80% of the time when it was wrong, while being *more* confident in their conclusions.

papers.ssrn.com/sol3/papers....

"Does the GOP think" 🤔

Bad

Pitt political science is admitting students but effectively the bare minimum number for a "full" cohort (down one from last year which was down several from previous years)

I know MB Neo must have been in dev for years, but it's so perfectly timed to steal Win 11 market share it's amazing

The US flailing about in complete disbelief that Iran would use their only viable strategic advantage strategically by closing the Strait should be an object lesson for other big countries considering invading smaller countries with outsized economic leverage

Source: Havana Syndrome investigation is "a massive CIA cover-up" | 60 Minutes YouTube video by 60 Minutes

Bonkers

My "the Department of D̶e̶f̶e̶n̶s̶e̶War is not at war" t-shirt has people asking a lot of questions already answered by my t-shirt

I'm real bullish on AI capabilities. It's humans I'm skeptical of: I'm really worried about careless use of AI which is basically all I've seen so far

Counterpoint: Claude will make mediocre scientists more productive and excellent scientists who want to do novel work relatively less productive

Can you say more? Do you think AI increases the value of in-person relative to MOOCs?

For example: I've reviewed complete jibberish papers now, submitted by serious scholars, who just didn't adequately vet whatever Claude gave them. If they can't do it, there's no hope for Joe B.A.

Now everyone can use Claude to be the next Avi Loeb...

I think this is correct but I think it will also make it much much harder for non-experts to assess expertise. I mean we already struggle with this ("OMG Dr. Oz says...") and it's only going to get worse. Convince your MBA boss that a solid PhD >> Claude 🤷🏼

Banana pudding is so much better than the sum of it's parts

"AI isn't ready to write a paper on its own yet - it's just not a very good writer. But I let it do all my coding and analysis for me it's great at that."

whichever part of writing a paper that you think AI can do better than humans is just the part of writing a paper that you are the worst at send tweet

The top scorer for Fri, Feb 20, 2026, UTC is: @benradford.bsky.social! They scored 360 points for #Wordle 1707 and solved it on row 1! That's better than 2,396 (~100%) other users. Never give up!

Wordle 1,707 1/6

🟩🟩🟩🟩🟩

Finally