Is everyone huffing paint?
Crypto guy claims to have built an LLM-based tool to detect errors in research papers; funded using its own cryptocurrency; will let coin holders choose what papers to go after; it's unvetted and a total black box—and Nature reports it as if it's a new protein structure.
Crypto guy claims to have built an LLM-based tool to detect errors in research papers; funded using its own cryptocurrency; will let coin holders choose what papers to go after; it's unvetted and a total black box—and Nature reports it as if it's a new protein structure.
Comments
If it's a half-decent proofreading tool authors might pay a buck or two (the quoted price) to have it glance over their manuscript - but to whom is that exciting?
https://pivot-to-ai.com/2025/02/22/google-co-scientist-ai-cracks-superbug-problem-in-two-days-because-it-had-been-fed-the-teams-previous-paper-with-the-answer-in-it/
Maybe someday in a couple of decades AI will do some of what they're hoping for. But not today. Crapto is crap.
These Broligarchs are seriously psycho.
So basically it's another QAnon style group...
The group was made up of "volunteers" that had an identical structure to QANON
The leaders ranged from a quack who claimed he could reverse the aging process to conservative influencers
"Decentralized Radical Autonomous Search Team Investigating COVID-19"
None of them had any actual relevant scientific knowledge or training... they simply claimed they were able to find the answers by "analyzing news reports".
Newsweek (which had been bought by a Right Wing group) claiming they had "broke the Wuhan Lab Story"
Bill Maher had a couple fake scientists on who repeated one of the group's entirely made up claims.
Vanity Fair gave them part of an article
All they had was made up claims of "reports" of WIV employees out sick with the flu in 2019 (but couldn't produce those reports)
and a claim that WIV cut off all of their data at the same time, which no actual scientist says happened.
Actual virologists cite that RaTG13 very well could have been a common ancestor of the COVID-19 virus. But that the genetic difference between the two represented 50 years of viral evolution.
https://bsky.app/profile/spinespresso.bsky.social/post/3lik646ftyc2c
“I thought, why don’t we go through all the papers?” Their AI tool has analysed 37,000 papers in 2 mo.
Only 37m to go (PubMed), so should be all done by 2192.
Someone please, please, please tell me that this overlapping Venn diagram of the gullible is a tiny, tiny part of the population.
Free column inches
Big clickbait
(This news came out early January)
PS: not medical advice!
Could see if a bigger LLM can crack it as being insane.
The hype train is running out of juice from the last infusion of useless junk and reaching for more.
Aristotle was smart, and "natural philosophy" was comically absurd. We're dealing with people who don't get that.
I'm guessing it pegged the "LOL" as English?
Or the author as an English speaker?
(My Dutch is just bad German, and my German's bad to start with, so I often use the translation feature.)
"It's funny. I was just curious."
There is almost no application of the science of evaluation in most AI evaluations today. It is a nascent field. The article hasn't spoken to enough relevant experts, IMO. Too insular. Why not talk to solid AI evaluation quants?
Why not also interview Arvind Narayanan, Eugene Yan, Nathan Lambert, Evan Miller, etc..
“AI tools are spotting errors in research papers: inside a growing movement”
Are they? These are unvetted black boxes. What evidence there is suggests they’re very bad at it.
It feels like repeating Elizabeth Holmes’s claims instead of asking for evidence.
Nature is wildly influential and reporting something like this is a choice. To me it feels like this situation with Science all over again.
The news story to me seems to be “Why does every new day bring a new unproven AI technology to fight fraud designed by someone who has never done science?”
But it doesn't sound like that is at all the intention here.
The article describes one, Black Spatula Project, exactly like what you said.
The other, YesNoError, involves cryptocurrency and sounds somewhat... odd.
Still, both have some initial promising numbers in the article. That's why Nature is reporting this.
And the rest of the article mentions in detail the limitations (false positives, etc.) of this approach. They don't sound naive. If it fails, it fails, but it's worth trying.
This is huffing lead microparticles.
https://bsky.app/profile/partickle.bsky.social/post/3ljsshpen6c25