Most insightful comment

This is already happening. Write code with an AI agent, agent creates PR, other agents review, coding agent iterates, PR gets merged, and hopefully real humans use the application and provide feedback

The latter step is also now staring to be replaced by AI (user directs agent to access app, get data, perform actions)

Similar thing with email

It isn't even partially human written, it's just an LLM's output.

I mean it's everywhere now. I complained already at my workplace multiple times, we get huge Claude generated documents, which people then ask Claude to summarize. I don't mind some of it in principle but people are too lazy now to input all the details so you get a bunch of fantasies that were explicitly ruled out already. It's tiring.

Sometimes I just write an email (in text) but sometimes it's a short LaTeX document (3 pages, title, date, my name, may have an introduction to ensure we use the same notation, usually no references because I'm lazy). Some of my coworkers are using AI that likes to make longer stuff and format it like a letter to the Nobel foundation.

Some are good, most are wrong or bad. Some are reproductions of others research work, some are early experiments, some are rediscovering the wheel, some are misguided ideas, and lucky a tiny part are good. Anyway, they are for internal use, we don't try to contribute to the thermal death of the universe uploading that crap to a free preprint server.

I don't understand:

a) Why they upload internal notes/reports to a preprint server?

b) Why are these reported as "academic papers"?

You add up all the claimed activities and then realize it’s more hours than are in a day so someone’s lying.

I don’t see how this author can claim that these papers didn’t “…suddenly materialize from a magical “chatbot” in 2026” when it’s obvious that 1-2 people are not producing 258 papers that are tens of pages each in 9 months with any sort of academic rigor.

If the papers were of a uniformly high standard then I might not have a problem with it.

If they are hoping some of them are good and the review process will filter out the bad then it seems like they are not doing research, they are writing hypotheses and at pushing the work of testing them onto the reviewers.

I could also see that being done to make some sort of statement about the review process, but it's not clear if that is their goal.

He's wasting people's time, which is already the resource that many people lack any to spare. I'm not sure if he realises this part.

When you compare that to the consequence of the hydroxycholoquine paper, which is comparable in public policy impact, where Didiee Raoult lost all its legitimacy and his job, it's easy to see that economics as an academic field doesn't value probity much.

Oh also at least 40% generated peer-reviews are contributing to this statistic.

"Show me the incentives and I will show you the outcome"

He was supposed to present a topic he learned about at the conference. Instead, he gave a cheeky talk which introduced the very concept of written language to the audience, as if we had never heard of it. Starting from the history of writing on tablets and so on, and ending in practical advice like, "when words are written down, you don't have to remember what they were exactly; you can just read the words again." I loved the chutzpah and could barely hold myself from chortling out loudly at some of his slides.

Why should anyone bother to think, if they can ask AI for a well researched, correct, and.. khm... "aligned" answer ?

Your intelligence and competencey are going to be 1:1 correlated to the quality and quantity of tokens a billionaire gives you access too (local, sure, they control the hardware markets as well).

Keep excercising your writing and reading muscles because it will definatley set you apart from the captive masses in the future.

I really don't like AI for document writing as a whole, especially within organizations. Code can be a form of communication, but for the most part it's an engineered thing that can be re-run, tested, verified and so on. But documents drive decision-making, and decision making is often one-time. Decisions should be studied and understood, and depriving authors of much of both is not having any good effects.