Ryan Carpenter

I bet ChatGPT loves to spit out meaningless percentages

Truly, the issue is not the AI, but the fact that we as a culture are allergic to very idea of non-dichotomous data

(I say this as one who just wasted a lot of time trying to find published epidemiological data on quantity of alcohol use -- no, prevalence of binge drinking does not count)

Headline that trumpets that half of teens use chatbots for schoolwork. In actuality, the survey found that 54% of students used a chatbot once in their lifetime for schoolwork.

Chart that shows that most teens do not usually use chatbots for schoolwork

Headline: CHATBOTS ARE TAKING OVER

Reality: Almost everybody used a chatbot one time and never again

OpenAI admits AI hallucinations are mathematically inevitable, not just engineering flaws In a landmark study, OpenAI researchers reveal that large language models will always produce plausible but false outputs, even with perfect data, due to fundamental statistical and computational limi...

OpenAI ”acknowledged in its own research that LLMs will always produce hallucinations due to fundamental mathematical constraints that cannot be solved through better engineering, marking a significant admission from one of the AI industry’s leading companies.”

You can’t trust chatbots.

NYT doing fine work posting these two stories right next to each other

Wait, though, that actually sounds pretty good

I remember when you asked a professor for a *strong* letter of recommendation

Now I guess you need to ask them if they will write you a letter without using AI

Mark Hyman is a man that is willing to claim that drinking milk causes cancer

Seven Symptoms Over Nearly 4,000 Days: Item-Level Variability in the Psychometric Properties of Daily Alcohol Use Disorder Symptoms in Young Adult Drinkers - Kevin M. King, Dahyeon Kang, Megan E. Schu... People experience symptoms of alcohol use disorders (AUD) in their daily lives, including more impairment-based symptoms (e.g., hazardous use, interpersonal pro...

I kept reading EMA studies claiming to test etiological theories of AUD, but I realized they never actually measured how people experience AUD in their daily lives. So we set out to see what we could learn with existing data.

1/19

journals.sagepub.com/doi/10.1177/...

Unconscionable.

Congress Is Reversing Trump’s Steep Budget Cuts to Science

This is a HUGE win…and one that happened because we ~collectively~ said “NO!”

But AAAS coming in and saying on record to the NYT “Science is doing ok. Things are not bad at all…” is baffling.

If things are hard for you as a scientist, please share in the comments.

www.nytimes.com/2026/01/10/s...

Also, the AAAS rep says that an overall drop of 4% for science funding is "pretty solid."

A 4% cut in funding is just about anything but "solid"

Clear-eyed, fact-based, and written to explain to normies. No “both sidesing.” No “Trump officials disagree.”
No gaslighting that what we can see with our own eyes might not be true.

This is journalism. Well done, @people.com

"This was a family that could've been like mine" -- Philip Bump breaks down crying on MS NOW when talking about the stuffed animals in Renee Good's car when she was killed

Imaginary food scene from Hook

Damn, and here I've been eating imaginary food this entire time

Ultraprocessed Foods Linked to Colorectal Cancer Risk in Women Under 50

After you listen to @michaelhobbes.bsky.social @yrfatfriend.bsky.social enumerate the many problems with UPFs as a scientific construct, read this NYT article and the linked JAMA study and see how many of them you can find (spoiler: it's all of them)

www.nytimes.com/2025/11/13/w...

Ultra-Processed Foods - Maintenance Phase Everyone agrees that processed foods are bad for you. When it comes to defining what they actually are, however, there is considerably less agreement. Support us:Hear bonus episodes on PatreonWat...

Just listened to the Maintenance Phase episode on ultraprocessed foods. I highly recommend. It is nutrient rich and dense in scientific fact, just like an all-natural, homemade, whole-food podcast should be

maintenancephase.buzzsprout.com/1411126/epis...

A 'Super Flu' Is Spiking in the U.S. — and Hitting Kids the Hardest Multiple states have reported child flu deaths in recent days, as cases and hospitalizations rise sharply across the country.

So as RFK ends annual flu vaccine recommendations for children, hospitalizations and deaths among children are rapidly accelerating--with prior data showing that nearly all the mortality is among those not vaccinated.
www.vice.com/en/article/a...

Today’s announcement that HHS is drastically altering the U.S. childhood vaccine schedule without a transparent process or clear scientific justification represents the latest reckless step in Secretary Kennedy’s assault on the national vaccine infrastructure.

Our statement: https://bit.ly/497Tj8V

Children will die with RFK Jr.'s new vaccine recommendations. We're NOT doing the "experts disagree" game. The US vaccine schedule was painstakingly constructed through a deep & thorough process by experts with unquestionable expertise. This is policy by fiat. It is pre-meditated murder. 1/

This is the energy I want to bring to 2026

(Full article: variety.com/2026/film/ne...)

“Be kind, be involved, believe in your art,” he said. “At a time when people tell you art is not important, that is always the prelude to fascism. When they tell you it doesn’t matter, when they tell you a fucking app can do art you say, if it’s that important, why the fuck do they want it so bad? The answer is because they think they can debase everything that makes us a little better, a little more human. And that, in my book, and in my life, includes monsters.”

Love this from Guillermo del Toro

Pretty Obvious 975 Leaked Memo on Leaks 729 Police Are Scrambling 918 Resolved and Unresolved 828

🎆🎇🎆2025 HEADLINE OF THE YEAR🎆🎇🎆
🎆🎇🎆🎇🎆🎇🎆GRAND FINAL🎆🎇🎆🎇🎆🎇🎆

After a seeding round and three rounds of voting, we have our finalists: Top seed Pretty Obvious will be facing off against 3rd seed Police Are Scrambling.

A reminder of how we got here:

My Open Letter To That Open Letter About AI In Writing And Publishing The tl;dr before you get into this post is this: the SFWA came out, said that some AI usage was okay enough in books for the authors of those books to not to be disqualified from winning a Nebula a…

Apparently I'm doing this -- my open letter to that open latter about AI in writing and publishing. I emerged from Hibernation Week to write it, so god only knows how much sense it makes. But at least a human wrote it, so that's nice.

Bonus: picture of my "cat," sweet baby Boomba.

I see the promise of AI here, the doors it could unlock.

But is AI the magic key? Or is it the key that's been determined most likely to work based on other keys someone saw once?

To say it without a tortured metaphor: I'm skeptical that AI can translate without putting its own stink on things.

I am in awe of anyone writing in a language other than their first. If tomorrow every major scientific journal started publishing in German, I would have to quit.

Your English may be imperfect, but I am in awe of it.

So I don't know exactly how I feel about all this. Depressed, mostly.

Despite what my colleague said, though, I don't think it's really a time issue.

I think a lot of it is a confidence issue. And I get it.

(I see this in my international students, too. And some of my native English speaking students, for that matter)

Why? My colleague offered two reasons:

1. Time. It takes a long time to write a review in a second language

2. To not sound super critical and hurt Americans' feelings 😂 (we love to dance around our words, Germans are direct)

My colleague felt it was appropriate to use AI for language corrections

Putting aside the hyperbole (sigh, can you even have AI without hyperbole?), I'm not surprised by the numbers.

Last month, I complained to my German co-author that a review we received seemed AI written. Their response: "Most of us non-native speakers do that now."

A table showing that most people use generative AI for peer review "never" or "rarely," with somewhat higher rates of use in places other than North America

This is from the Frontiers report the article covers (Nature is so nice to help Frontiers spread the good news of AI!)

Only 16% of North American researchers report regular AI peer review use. Almost nobody reports frequent AI (Nature forgot to mention that), but it's more common in other regions