Friends, you may have seen this but if not, sharing:
"The LLM-generated proof hinges on a particular mathematical argument that it presented as its own but that actually first appeared in a 2016 paper by Miller and a collaborator."
14
Replies
19
Boosts
@timnitGebru yet again, the operators of these things lack the means to evaluate their outputs.
Love the graphic.
Can't read the article as I am burnt out on any hype on AI.
@timnitGebru that site blocks my vpn wtf
@timnitGebru Who could possibly have predicted this? 🤦
@timnitGebru OpenAI is a bunch of no-talent ass clowns.
@timnitGebru The one thing you can't claim about an LLM is that it hadn't seen the prior art.
"Two of the most exciting results, the experts say, incorporate preexisting ideas from the recent mathematical literature without properly citing them. This contradicts OpenAI’s initial press release, which said that the problems Astra addressed “have been open and seen no progress on the main result for at least a decade.” (OpenAI has since updated the language to be more accurate).
@timnitGebru define “properly”, is it mentioning the 10 zwar old paper or not?
“They are running roughshod over the work of others who came before them in a deliberate way,” says Stephen Miller, a mathematician at Yeshiva University, who argues that OpenAI has effectively plagiarized his own research. “It seems completely systematic to me, and it points to research misconduct.”
"The LLM-generated proof hinges on a particular mathematical argument that it presented as its own but that actually first appeared in a 2016 paper by Miller and a collaborator..."
"Like a number of recent AI breakthroughs, it pasted together ideas from the mathematical literature to build a new theorem. Once again, the LLM’s trick is its superhuman patience for assembling puzzle pieces, not the ability to make some profound intellectual leap."
"“there is the big PR machine that wants to sound as impressive as possible and does not care about being 100 percent accurate,” he says."
@timnitGebru It's particularly frustrating that the academic misconduct is most likely completely unnecessary to OpenAI's goals. First, the LLM is most likely capable of providing the correct references if you prompt it to do so (at least that's what Ive heard, from people who have received neat proofs from an LLM and have asked it to cite instances in the literature where that argument was used before). Second, adequately giving credit to people doesn't make novel proofs of conjectures any less exciting to mathematicians; doing it right would serve their PR purposes equally well.
@oantolin I don't think it would serve their PR purposes because it would look more like the machines that win at chess or other approaches that look more like brute force.
"The discovery stunned Francesco Fournier-Facio, a mathematician at the University of Cambridge, who studies group theory—at least until he “engaged with this breakthrough as I would if a human had written it,” he says. The result, he and some of his colleagues found, wasn’t as novel as it first appeared."
@timnitGebru I was imagining OpenAI just removing the claim that no progress has been in made on those problems in a decade, getting ChatGPT to reference the literature appropriately, and still saying "here are 10 conjectures mathematicians care about that our new model solved". I think that would be equally good for them as PR and they wouldn't have faced some backlash for not crediting people whose work is used.
If OpenAI said "our models merely combined some standard techniques in the area, and thus don't really show our model is super smart but rather that these conjectures were lower hanging fruit than previously thought", then I could see that people would feel this more like a brute force approach. But the company wouldn't have to say that, they could just properly cite people's work, only claim to have solved these conjectures (which at least for a few I've heard about from experts, does seem to be the case). Probably in that alternate history mathematicians would have started explaining that the proofs don't really have some astonishing new insight out of the blue, but only combine some know results in a new way (like what Fournier-Facio said about the non-sophic group example), but that wouldn't have really hurt OpenAI's PR effort since nobody pays much attention to mathematicians anyway!
