"Dear Diary, had an enjoyable day stirring up my ignorant base by pretending that migrants on a Spanish territory in North Africa will somehow be arriving in Dover any time now"

Yes, I think so. What’s most notable about LM’s is that they’re at least claimed to be able to contribute throughout the entire research process and thus the timing of where they contribute most influences the way in which they shift research effort.

Does this result generalize to _any_ productivity-enhancing technology*? It reduces the friction of some component of the research process, and the "most productive" style of project moves somewhere else in this landscape. * again under the assumption that it works

So we need metrics that specifically penalise “more but worse” in output. E.g. lab 1 produces 10x more papers than lab 2 but the papers from lab 2 are cited 10x more. 2👍 Or: lab 1 produced 100 papers and all of them were cited below the expected median for the journals in which they appeared. Fail!

If so, set that to zero in the model and apply the other results.

Very interesting stuff... I'm not really sure how much science is done in a way that conforms to your model though. I'd argue that in e.g. biomedical research a lot of researchers are doing incremental work and don't really have a "discovery period".

12. Here's the key insight though: as labor-augmenting technologies LLMs make researchers more productive and therefore they *raise the opportunity cost of researcher time*. As a result researchers will have incentives to invest less time in polishing a current project before moving on to a new one.

11. By contrast, in empirical disciplines with a heavy lab or fieldwork component that cannot be handed off to LLMs, and where instead these models are most useful in writing up results, researchers will become less selective.

10. In theoretical disciplines where LLMs can hypothetically aid researchers in screening for the most promising projects, researchers will become more selective about what they publish.

9. We present a detailed comparative static analysis of this model. We find that the use of LLMs to accelerate workflow will influence how selective researchers will be in taking projects to completion.

8. We imagine the scientists taking their work through the following lifecycle. Initially the value of a project is unknown. After a discovery period, the value is revealed. Scientists then can abandon the project and start over, or choose how thoroughly to develop it before publication.

7. In our model, we focus on how the availability of labor-augmenting technology shifts (i) which projects scientists choose to take to completion, and (ii) how thoroughly they complete any given research investigation before publishing it and moving on to something new.

6. Our approach is to adapt a classic biological model in optimal foraging theory—Eric Charnov's beautiful marginal value theorem—to explore how scientists allocate their labor across projects. Instead of gobbling blackberries or whatever, scientists develop projects and eventually move on.

5. Even if they were to perform the tasks allocated to them flawlessly, LLMs would disrupt the balance of frictions and inducements that scientific norms and institutions impose, reshaping scientific practice and the allocation of cognitive labor. This is classic "New Economics of Science" stuff.

4. To distinguish the effects of flawed technology from the general consequences of labor-augmenting machines, we model LLMs as imagined by their most enthusiastic proponents: capable of reducing time-costs without increasing errors and mistakes, at negligible financial cost.

3. Indeed, LLMs will change the ways that we scientists spend our time. That horse has left the barn. But in this paper, we reach the sobering conclusion that LLMs are not likely to fulfill our fantasies of freeing up more time for leisurely contemplation or more thorough investigations of nature.

2. When I express concern about how LLMs are changing the way that we do science, people often say "No, it's be great. AI will do all the boring bullshit and tedious drudge work. We'll finally have time to read broadly and think deeply."

1. We—@eduede.bsky.social, @mjcrockett.bsky.social, Kevin Gross, and I—have a new preprint on the arXiv today, based on ideas that emerged during an @sfiscience.bsky.social workshop in November 2024: The unintended consequences of large language models as a labor-augmenting technology in science.

The unintended consequences of large language models as a labor-augmenting technology in science

As a labor-augmenting technology, large language models (LLMs) have the potential to accelerate scientific activity across the research pipeline. But even if LLMs perform on par with human experts at ...

arxiv.org

13. Hence our dismal conclusion: rather than finding that LLMs free us to do a better job of what we were doing before they came along, they shift scientific incentives (and the playing field of academic competition) in ways that compel us to do more and more, faster and faster, less and less well.

If this is what happens to a scientific field as a whole when LLMs dominate the workflows of researchers, I can only imagine that the outlook is even worse for any individual researcher who is soldiering on without using them. 😞

Carl T. Bergstrom@carlbergstrom.com · 2w ago

13. Hence our dismal conclusion: rather than finding that LLMs free us to do a better job of what we were doing before they came along, they shift scientific incentives (and the playing field of academic competition) in ways that compel us to do more and more, faster and faster, less and less well.

New data reveals that enrollment in California's state Medicaid program is plummeting as Trump's big ugly cuts take effect. But there's a way to reverse this trend: pass a one-time, 5% tax on California's billionaires.

Why I Support The California Billionaire Tax

Billionaires are crying foul about California's proposed wealth tax.Don't fall for their fearmongering.A one-time, 5% ta

youtube.com

I’ve been yelling about it this all year. Municipalities still act like these are unfortunate incidents instead of what they are. They’re attacks on critical infrastructure. We’ve underinvested in modernizing essential systems and infrastructure while degrading cybersecurity.

Jeanne Gibson, beloved “Rosie the Riveter,” passed away on July 29, 2026 at 100. Gibson became a welder at Todd Pacific Shipyards in Seattle, helping build destroyers for the Allied war effort during WWII. Gibson helped pave the way for millions of women who stepped into critical wartime roles.

Don't forget that Audible is owned by Amazon and you should use @libro.fm instead, which allows you to support an indie bookstore in the process

Buy audiobooks & support local bookstores

Libro.fm makes it possible for you to buy audiobooks directly through local bookstores.

libro.fm

I work at a public library. People who want to use AI seem to have no trouble doing so. People who don't want to use AI have a harder time. I crowdsourced and then finalized this document "How to disable or avoid intrusive AI" and then gave it a short URL: notoai.org Please pass along if it's useful

What could possibly go wrong? AI autonomously designing nuclear reactors. Aren't out-of-distribution events a big deal for nuclear reactors? www.science.org/content/arti...

……

Michael Clemens@mclem.org · 6d ago

I wrote last year about the US Administration's plans to eviscerate Optional Practical Training, America's largest single source of global STEM talent: www.piie.com/blogs/realti... That came to pass. The White House is sharpening its tools to shatter OPT, and thus kneecap many US universities.

There are so many thoughts. Like - why are governments cutting support for basic and crop research? Why are we allowing energy-demanding, water-hoarding "data centres" to be built? When will people wake up to the future we are allowing to happen? ☹️ www.theguardian.com/environment/...

‘It’s too fast, too big’: speed of climate crisis is devastating crops, say farmers

Drought and extreme heat are killing plants, driving up prices while also making some farms economically unviable

theguardian.com

Some nice aquarium sci-comm

I have never seen a stuffy with a more kickable face