Sean Mackinnon

You forgot also the [does not do X, but coefficients differ only by 0.01 when they do X]

Uni Admin: The numbers must go up.
Me: So, increased scholarships and residence spots?
Uni Admin: No, not those numbers

Yeah, I expect something similar. The current systems are calcified in a way that will be hard to change quickly. The first lever is always "make it harder to publish anything"

AI papers reviewed by AI are totally going to be a parallel thing to people doing science a more human-centric way.

I think that the obvious system change that will need to occur is that all the various selection processes that use number of publications needs to move to "Please submit only N of your best publications for review" where N is a small number.

h-index and pub counting are going to become obsolete

Picture of dow at 47501

😬

Excited, haven't played yet but playing this game is my weekend plan!!

Load-bearing note right there

Simpsons bed meme: Milhouse's dad: "I consume ~40 million tokens across ~20 presumably high-end agents every one minute and 54 seconds." homer: "I install programs with an .exe file"

"Even with ZDR arrangements in place, Anthropic may retain data where required by law or to combat Usage Policy violations and malicious uses of Anthropic's platform. As a result, if a chat or session is flagged for such a violation, Anthropic may retain inputs and outputs for up to 2 years."

I would pretty closely read Anthropic's guidelines on ZDR arrangements. I'm not confident that it'd be safe for truly sensitive data, especially any data that the US government law might permit requesting (e.g., Patriot Act). Not sure it'd be viable in Canada.

docs.posit.co/posit-ai/use...

Like, could you just imagine in the pre-AI days if someone was like "I hired my assistant to roleplay as George Washington, and asked their opinion on the Iraq war. This is what they said" you would have thought they were a real nutter

I legitimately laughed at proposing multiple imputation plus last observation carryforward

I had to look up shark biting an internet cable to figure out what the bottom one was lol

What we have discovered is that Claude can indeed produce engagement bait that people on Bluesky cannot resist

If there is one thing we love in Atlantic Canada, it is cheap, ubiquitous, mid-quality coffee with extremely high sugar content

Truly an all-time best. I still think about this one sometimes.

I wonder if that manager knows how infamous he is in statistics circles lmao

Ohh, very cool thank you!! I will pass this on for sure

Sorry, I shouldn't doom on your timeline!!

Had a brief moment of despair there. Only way to survive is to stick to your morals and keep trying. Just wish I felt like I was on the winning side for once

It's not that I can't see that AI agents can do so many incredible things, I just can't see past the social ramifications of relying on them

The AI companies just have so much capital they can just undercut so much on price, quality doesn't even matter. People and quality of life don't matter.

Ah man, I shouldn't have read that substack post, honestly really stresses me out

I want a future for my students where their very thoughts aren't controlled by corporations and where they can still get good jobs like mine.

The social trajectory right now is just so bleak

Oh wow, thank you for all of this! Of all this stuff, which do you think is most beginner friendly? It is for a grad student who knows the general linear model in a frequentist framework, done some traditional EFA before, but never done Bayesian analysis of any sort before.

Does anyone have any resources/readings that could help (or advice on whether this is a good fit)?

Intuitively, it seems like a Bayesian approach is a better epistemological fit for indigenous statistics, but Bayesian analysis is something of a weak point for me. Would appreciate any thoughts! 2/2

Calling all Bayes #stats people, I had a student who wants to do exploratory factor analysis in a Bayesian framework. They are working with indigenous people and developing a scale, and want to incorporate prior knowledge about the factor structure & factor loadings. 1/2

You joke, but those of us non-AI users are basically becoming artisanal manufacturers at this rate. Like, we are gonna be like the handcrafted artisan pastries vs. a Tim Hortons donut.

I mean, based on the dictionary definition "to want greatly" or "to yearn for" seems like their claim is wrong, at least.

I mean, a whole streaming service called themselves Crave, presumably not because they expected everyone to watch reruns they saw before

Yeah I couldn't place the sound, but me too. Pretty loud wow

Fascinating, reminds me a lot of Twitch Plays Pokemon

Reading through this page and uh, wow. Getting hard to tell the joke posts from reality, huh?

Hard to believe someone created a way for AI agents to play Sim City and yet

I really unironically love this movie, one of my comfort watches

Me too, tbh