Berna Devezer

OK last one because this hits close to home Not only is it a great misconception that replication = truth (one doesn't imply the other) but also replicability is not a reliable metric to establish a crisis either (see our paper). arxiv.org/abs/2604.26268

The Difference Between "Replicable" and "Not replicable" is not Itself Scientifically Replicable arxiv.org

ivy full prof who publishes 25 papers a year calls for academics to publish less

washingtonpost.com

I was just reminded of this old gem, which seems to capture the kind of degenerative process you're referring to

rpgroup.caltech.edu

abstract: We show that fully autonomous AI analysts built on LLMs can, cheaply and at scale, produce the analytic dispersion observed in human many-analyst studies limitations: the analytical paths AI analysts explore need not coincide with those human analysts would traverse 🤔

pnas.org

earlier this week I sat at a meeting where an admin showed us AI-generated slides, stopped on a slide that featured an exceptionally cringe and cryptic AI-generated image, and read a very long AI-generated poem about the administrative issue in question. no one should ever have to sit through this

Beyond "It's not just X, but also Y" construction, I find AI-produced corporate speak very cringe. This is from a popular brand value report: "In a sea of intelligence, wisdom is the helm - and brand is where it lives." Not that corporate speak isn't cringe without AI but this is obnoxious.

(seemingly) everyone loves to dunk on AI-generated slop and condemn those responsible in no uncertain terms on here. yet people keep reposting pure falsehoods, baseless speculations, intentionally misleading takes to my timeline. we should try posturing less and reflecting more sometime.

been seeing an ai paper circulated intensely with the caption "Harvard just dropped this study" for a phd candidate's job market paper. it's bothering me. feel like we should be centering the student instead of the prestige of the institution. if the research is good, maybe also focus on its merits.

Tired: Wish there was an edit button to fix typos Wired: Typos are all mine; 100% organic, human-generated Inspired: I'll sprinkle my LLM-generated text with typos to signal authenticity Retired: I'm never writing or engaging with written word ever again

Totally vibing here, but hope we'll soon get collectively fed up with llm-speak and refuse to entertain it. Getting impossible to seriously engage with extensive text without an underlying thought in interpersonal comms. It's introducing emotional barriers and negative judgment. Who needs that?

I provided detailed AI-use guidelines for the project in my undergrad class, and I can report that there was no noticeable sign of unauthorized AI use in presentations or reports. Based on other accounts on here, I feel that this is now the exception? My students pretty much followed my guidance.

I truly don't get the point of peer reviewers using AI to review a paper. If the authors need generic comments or simple error checks, they can use AI themselves. If the journal/editor needs it, they can do the same. What is the point of asking a peer reviewer to ask AI? Setting all other objections

The sentiment in the classroom wrt AI has changed drastically since last spring. Suddenly, extreme enthusiasm is replaced with skepticism, distrust, need for clear guidance. Have we passed the peak of the hype cycle? Is that it? Students are smart(er than uni admin)! They care abt the right things.

Wish some of these efforts to relentlessly inject AI in every step of science were at least a little interested in and informed by the history of science. Feels like people don't even wanna learn about the thing they wanna save. They wanna be saviors for free! Such a lack of curiosity depresses me.