Josh McCrain

this'll be an interesting rorschach test

my guess is next time i teach intro (up to 400+ students) it'll be 30% of the students with an accomodation

some insane numbers in this story.

i asked my office of disability last year when teaching a 420 person class what assistance they provided in meeting each individual accomodation. you can guess the answer.

www.theatlantic.com/magazine/202...

more broadly, sure untenured faculty teach undergraduates in this environment?

The implication here is that he put the national guard into a dangerous position (ie a big scary city) , which I think is counter the point wanting to be made

absolutely deranged. why use a calculator when you can simply do the OLS matrix algebra by hand? same argument

Yes, exactly! Anybody refusing to detail this kind of use case to students is doing them a disservice. Would be akin to refusing to use a personal computer when they became prevalent

dude! this is awesome, where are you finding this

Bluesky has tons of diversity in viewpoints. You've got the crazy left, the progressive left, and the center left

cool paper. yet another way policing is similar to academia

Oh my god there's like 25 coauthors on this

Incredible. Good thing nobody trusts JAMA *touches ear* oh shit

Lol

I think low, right now. But agent based browsers will easily be able to do this and they are very close to being used at scale

Say more?

Some good stuff from Sean here:

what's this from?

yep I think this is probably where we're headed. all of this will of course make survey work much more expensive

I can also vouch for prolific as a presently very good sample provider. The paper is outlining a threat that is surely not realized yet

Yes but I think the core concern is it's essentially impossible to know *how much* AI has taken over platforms

the scale fully depends on how you throttle, i guess. but yes, all valid points. but the "cost" to making a bot is extremely low in terms of time, and this will get increasingly easier moving forward especially as agent-based AIs become more prevalent

it's definitely real. link to full draft in replies. possible sean broke the embargo early at PNAS

bsky.app/profile/anth...

To me, the point of this paper though is that we cannot know when the pool gets "bad" because responses look like normal quality human response

Trivial to program llms around both

Correct, but it will only get easier and as he points out economic incentives are there

I love prolific and have spent thousands with yall (didn't see your profile), but the key is not identify verification. It's that once identified, it's impossible to determine whether they're humans doing the actual responding. Throttling seems to be the best approach, though

Everything prolific does there can be fooled as Sean outlines

Definitely the worst response to this by far is "why would you study public opinion anyway just talk to people" I mean jfc

Another obvious point here is that it also invalidates attention checks....