iiinova — where your digital self keeps evolving

See, now this I can get behind.

I've actually done this out of genuine curiosity lol. Always interesting to know the interpretation your paper will have among those who can't read

220

Robin Blythe @rbly.bsky.social · 1h

I often wonder what role these guys think they would have had in a Nazi-run world. The entire MO of fascism is to continuously find outgroups to vilify; their weird fetishes probably put them in one of those outgroups sooner rather than later

010

Robin Blythe @rbly.bsky.social · 2h

Are there any good reasons for people to take photos of slides sharing yet-unpublished research at conferences? I've now witnessed two cases of research theft from colleagues via this behaviour and can no longer assume people who photo your slides do so in good faith. Too uncharitable?

static.klipy.com

Raccoon Steals Divine Orb in Poe Path Of Exile

ALT: Raccoon Steals Divine Orb in Poe Path Of Exile

100

Robin Blythe @rbly.bsky.social · 7h

Few things more terrifying than overhearing a med student say "who cares bro, just ask Claude" regarding the curriculum

080

Robin Blythe @rbly.bsky.social · 17h

For those of us without talent, alas, the cruel hand of fate has other ideas

010

Robin Blythe @rbly.bsky.social · 07/10/2026

Seems ok to me if we shunt most people who come to uni solely for job training over to trade schools

020

Robin Blythe @rbly.bsky.social · 07/10/2026

I do miss having conversations about other things at work. It's reached shibboleth status, you have to say it in every convo.

020

Robin Blythe @rbly.bsky.social · 07/10/2026

To be fair to Calum, the man is a pie artiste. Weird crossover to Japanese men's fashion, but as a chef he has been extremely influential!

1280

Robin Blythe @rbly.bsky.social · 07/10/2026

In a way I'm sort of glad. The older process never worked all that well, and maybe we'll be forced to figure out something better.

110

Robin Blythe @rbly.bsky.social · 07/10/2026

My favourite is when the dataset comes with all the rows with missing data dropped so that it is "cleaned" for you 🥰

020

Robin Blythe @rbly.bsky.social · 07/10/2026

This feels like it's been the case for decades, really. If you're still cold applying for jobs as a professional, someone in your life needs to take you aside and give you some real talk

250

Robin Blythe @rbly.bsky.social · 07/10/2026

I think I'm in the same boat as you. I have a strongly defined research q and know in principle how to answer it, but LLM back-and-forth usually gives me a couple of "huh, didn't think I could do that" moments which are illuminating. I think they make my work better.

010

Robin Blythe @rbly.bsky.social · 07/10/2026

Somehow I've never heard of this, guess my circles were too health services research focused. Thanks for letting me know about it, sounds excellent!

000

Robin Blythe @rbly.bsky.social · 06/10/2026

Good thread in the OP, and nice follow-up too. What irks me most about the degrowth movement is that it's a shotgun blast straight to the kneecaps of basically every developing economy in the world. "Yeah, no growth/living standards for you, we're at capacity."

120

Robin Blythe @rbly.bsky.social · 06/10/2026

The recent variation I've seen replaces 4-6 with clearly LLM-coded IPTW then claims the results are ✨causal✨ i.e., Babe wake up, new p-hacking strategy just dropped

050

Robin Blythe @rbly.bsky.social · 06/10/2026

Another nice blog post (not just because I agree with it!) that reinforces a particularly tough pill for many scientists to swallow: there are lots of really strange assumptions that do the heavy lifting when we seek to call our process "exploratory", often at the expense of theoretical grounding!

031

Robin Blythe @rbly.bsky.social · 06/10/2026

I'm absolutely sure that containerisation was invented solely due to anal sysadmins

120

Robin Blythe @rbly.bsky.social · 06/10/2026

Slight tangent but yeah, correct. I actually went into this a little bit (indirectly) in that preprint I shared last month. Basically, due to the confounding effect of treatment -- I.e., nobody goes untreated -- you have to look at process outcomes.

010

Robin Blythe @rbly.bsky.social · 06/10/2026

Maybe a hot take, but IMO the best thing about propensity scoring is that it makes clear how absurdly difficult it is to reduce complex patients/clinician+patient interactions down to a handful of variables. It virtually always obviates some missing and/or irreducible context. #statsky #episky

151

Robin Blythe @rbly.bsky.social · 06/10/2026

Hmm, not sure that's going away any time soon. Academic writing is largely a rhetorical exercise, not a proof where you can lay out the steps and put QED at the end. It is only as effective as the context in which it's delivered. Might as well cut out the paper entirely and just publish the code.

050

Robin Blythe @rbly.bsky.social · 06/10/2026

This is handy! Though usually I forget I even have MASS loaded

000

Robin Blythe @rbly.bsky.social · 06/10/2026

Wow! I'm almost jealous of the snail mail rejections because at least you can blame the post for taking too long. Waiting on slow reviewers in the age of instant reply is possibly worse.

010

Robin Blythe @rbly.bsky.social · 06/10/2026

Happened to me during a live demo after I made some last minute changes before the presentation. Embarrassing to say the least!

120

Robin Blythe @rbly.bsky.social · 05/10/2026

Every time I forget to load tidyverse last

3230

Robin Blythe @rbly.bsky.social · 05/10/2026

Yeah I agree. Most of the discussion I've seen has been from people annoyed that reviewers are checking out and just handing over completely to LLMs. That said, it would be a nightmare trying to integrate the cache from individual users' past interactions for any reproducible analysis

110

Robin Blythe @rbly.bsky.social · 05/10/2026

Because the discontent about LLM use by reviewers is not driven by questions of model performance...

100

Robin Blythe @rbly.bsky.social · 05/10/2026

If you're asking a reviewer for comments, then you want the reviewer's expertise, not the LLMs. Journals can/should do an internal review by LLM if desired, but a reviewer passing off the output as their own is disingenuous and not the point of peer review.

100

Robin Blythe @rbly.bsky.social · 04/10/2026

I've always wanted to fit a GAM and collaborators give me this look, so I don't bother. One day!

static.klipy.com

Dog Side Eye Rolling Window Up

ALT: Dog Side Eye Rolling Window Up

040

Robin Blythe @rbly.bsky.social · 02/10/2026

Do these make up such a big share of the sluggish growth? I assumed it was more intensive work like manufacturing that got wrecked as a result of tariffs, rather than anything to do with AI. There is existential AI-related unemployment fear that they can use, though.

100

Robin Blythe @rbly.bsky.social · 02/10/2026

This isn't my area of expertise but I bet these data can be dug up somewhere. An ex-colleague of mine used to work on tracking child trafficking over eBay, maybe using similar methods. I'd love to see someone take Meta/eBay to task for this.

010

Robin Blythe @rbly.bsky.social · 02/10/2026

That is extremely interesting if true. Has there been any research on it?

100

Robin Blythe @rbly.bsky.social · 02/10/2026

Are you sure they weren't a little unstable in the first place? A lot of people in this space have always been socially awkward, now all of a sudden they're being given full license to behave however they want, and paid unlimited money to do so. That has to count for something

040

Robin Blythe @rbly.bsky.social · 02/10/2026

{patchwork} and multi-panel plots are your friends, I can sneak in like double the illustrations usually

000

Robin Blythe @rbly.bsky.social · 02/10/2026

They tell me that they find the biggest associations, the best associations. Nobody knows those associations better than me. People come up to me - big, strong guys - and they say, sir, your causal inference is amazing. It's true. We use causal inference to find the most beautiful associations.

170

Robin Blythe @rbly.bsky.social · 02/10/2026

In health, senior scientists get invited to be on like 2 a month. It's kind of a joke, usually just dredging for associations.

150

Robin Blythe @rbly.bsky.social · 01/10/2026

This is how I know a lot of my colleagues are using AI to write their papers. Rather than craft an argument in which you build to a conclusion over several paragraphs or even pages, it ends each paragraph in one. It makes for a tiring and weak read.

040

Robin Blythe @rbly.bsky.social · 01/10/2026

Honestly, same here. No room for ambiguity.

000

Robin Blythe @rbly.bsky.social · 01/10/2026

This makes me feel like I need a degree in CS just to run lm(). Rams home the 'standing on the shoulders of giants' thing, except it's giants all the way down!

110

Robin Blythe @rbly.bsky.social · 01/10/2026

Yeah this is the biggest counterargument. There is no such thing as an airtight prompt, the output is a snowflake. Basically always just snow, but with an incomprehensible mechanism creating it.

010

Robin Blythe @rbly.bsky.social · 01/10/2026

Absolutely -- I think the biggest positive externality from learning to code is that you have to also reverse-engineer it and improve your mechanistic knowledge in the process.

010

Robin Blythe @rbly.bsky.social · 01/10/2026

I think this is probably the most important counterargument to me. If LLMs and other human -> computer languages are built on stochastic models, they are basically entire irreproducible.

010

Robin Blythe @rbly.bsky.social · 01/10/2026

A follow-up to this is that I will acknowledge that people can already fit garbage models, and coding at least provides necessary friction. But eventually I think a code interface can also identify these issues and point out why the model is crap.

010

Robin Blythe @rbly.bsky.social · 01/10/2026

These are both fair. Consider someone with good theoretical knowledge who can describe the analysis specifically. Isn't the code just an intermediary language step? It's distinct from knowing what is feasible and appropriate.

200

Robin Blythe @rbly.bsky.social · 01/10/2026

These are all still true! You should always need to specify exactly what you would like to do. But I think, increasingly, there will be less friction in this process in which minor syntactic errors are ironed out. If you can describe it fully in lay language, isn't that good enough?

110

Robin Blythe @rbly.bsky.social · 01/10/2026

Weirdly, this was the post that spurred this particular line of thought

010

Robin Blythe @rbly.bsky.social · 01/10/2026

Complaining about people not knowing how to code seems a little weird when I guarantee most people have absolutely no clue what, for example, glmer() is actually doing under the hood

440

Robin Blythe @rbly.bsky.social · 01/10/2026

Given that coding is basically a mapping from human-to-computer readable language, it really does seem like the natural evolution is to abstract away the human-readable part (ie, code), making it a smooth transition from human lay language straight to binary.

230

Robin Blythe @rbly.bsky.social · 01/10/2026

"Climate change: What does it mean for your summer home in the Blue Mountains?"

010

Robin Blythe @rbly.bsky.social · 01/10/2026

Not only that, renewables got a huge shot in the arm from a certain, particularly idiotic decision to close the world's largest shipping route for oil!

040