"This role will develop LLM-based scientists to engineer materials at the atomic scale. "OH MYGODNO
-
@aud they'll select a few examples to cherrypick, post a paper, get funding over real scientists, do absolutely nothing, then move on to the next field to colonize
-
Asta [AMP]replied to pedestrian cyclist last edited by
@[email protected] the LLM scientist just keeps saying: "one word: nanomachines, son" no matter what we throw at it!
-
Asta [AMP]replied to d@nny "disc@" mc² last edited by
@[email protected] "our LLM scientist, trained heavily on a corpus of the data that Aaron Swartz was killed for trying to give away, almost perfectly reproduced a well known, well cited, well quoted result after only 10,000 trial runs!"
-
d@nny "disc@" mc²replied to Asta [AMP] last edited by
@aud don't forget we specifically optimized parameters to make it generate those results in a manner exactly like standard p-hacking but what is the reproducibility crisis when there's science fiction we can appeal to
-
-
d@nny "disc@" mc²replied to d@nny "disc@" mc² last edited by
@aud also see "temperature" https://circumstances.run/@hipsterelectron/113314567769180333
-
Asta [AMP]replied to d@nny "disc@" mc² last edited by
@[email protected] (this is such a good post)
-
Asta [AMP]replied to d@nny "disc@" mc² last edited by
@[email protected] yeah, I definitely saw that, and I agree.
(unrelatedly, as someone who actually studied statistical thermodynamics, complex systems, sampling problems, protein binding, and has had a career in HPC and done ML, I really hate when AI fucks dismiss me on technical points (that I'm correct on). The fact that they use temperature reminded me of this). -
@[email protected] (but then, of course, I watch how these people dismiss Dr. Margaret Bender and Dr. Timnit Gebru all the time, people who are 100% specifically experts in this field and I'm like, right.)
-
@aud "tell me you ate up a deepmind PR stunt and didn't read the follow-up literature without telling me"
-
@[email protected] "why should I read the literature when the investors sure as hell don't"
-
@[email protected] you know, I posted that as a joke, but I dunno if I can really disagree to a certain extent. It's not like the person writing the description gets any say in what the company thinks it wants.
-
@aud taking your point even further: the publications can't even be bothered to read the goddamned things anymore
(because they don't pay for almost all of the labor that creates their value)
-
@[email protected] sweet fucking racket, being a journal
Although since the current trend in "AI" is "publish" to arXiv and then who gives a shit if it ever even gets peer reviewed, I guess the only sucker here is us.
