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..."
"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."
-
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."
@timnitGebru To be fair, "OpenAI’s latest math breakthroughs commit research misconduct, experts say" is incorrect.
It's the people using it who did; they're inferring intent where there is none, giving AI attributes it doesn't possess.
-
@timnitGebru To be fair, "OpenAI’s latest math breakthroughs commit research misconduct, experts say" is incorrect.
It's the people using it who did; they're inferring intent where there is none, giving AI attributes it doesn't possess.
@timnitGebru Sorry about teaching my Grandma to suck eggs and all that 🤪
-
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."
@timnitGebru
AI did not do "academic misconduct", that's impossible. It's not an author.Academic misconduct happened when people submitted work that wasn't theirs
-
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."
Is this not how any LLM would “solve” any math problem? It can probabilistically generate combinations of text from its training set until one particular combination accidentally overlaps with an actual proof?
-
Is this not how any LLM would “solve” any math problem? It can probabilistically generate combinations of text from its training set until one particular combination accidentally overlaps with an actual proof?
@HeavenlyPossum@kolektiva.social @timnitGebru Correct. Monkeys typing would achieve the same result, but it would require an enormous number of monkeys and take an eon.
-
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."
@timnitGebru The one thing you can't claim about an LLM is that it hadn't seen the prior art.
-
S sebastian@social.itu.dk shared this topic
-
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."
@timnitGebru OpenAI is a bunch of no-talent ass clowns.
-
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."
I have seen very smart people be fooled into thinking that LLMs can reason.
I’ve tried to show them that this is not the case but they don’t see it.
My suspicion is that they are privileging first order logic and logical semantics as superior to ordinary language and so assume that these are not LLM-bullshitable.
When I’ve tried to point out that there is no such thing as a pure language of meaning, immune to bullshit, as Frege and Russell hoped for, they just can’t believe me.
-
Is this not how any LLM would “solve” any math problem? It can probabilistically generate combinations of text from its training set until one particular combination accidentally overlaps with an actual proof?
@HeavenlyPossum @timnitGebru yeah that's my understanding too. The programs have no sense of logic or truth. But I'm not an expert and I also don't care because they suck on all fronts anyway.
-
F folfdk@helvede.net shared this topic
-
"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.
-
@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 @timnitGebru it makes it less exciting to non-experts, which there are a lot more of.
-
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."
@timnitGebru Who could possibly have predicted this?
-
@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."
-
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."
@timnitGebru that site blocks my vpn wtf
-
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."
Love the graphic.
Can't read the article as I am burnt out on any hype on AI. -
I have seen very smart people be fooled into thinking that LLMs can reason.
I’ve tried to show them that this is not the case but they don’t see it.
My suspicion is that they are privileging first order logic and logical semantics as superior to ordinary language and so assume that these are not LLM-bullshitable.
When I’ve tried to point out that there is no such thing as a pure language of meaning, immune to bullshit, as Frege and Russell hoped for, they just can’t believe me.
@CptSuperlative @timnitGebru #AI is #clankers
all the way down, and that’s precisely why wankers love 
it *so much!* -
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."
@timnitGebru yet again, the operators of these things lack the means to evaluate their outputs.
-
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."
@timnitGebru
That's extremely weak. Almost absurd.
Of course llms would build on exciting work. We all do that. Saying they did nothing just because they "puzzeled a lot of pieces together" is laughable. Why didn't anyone else do that already!!?
And if some human did and solved any of these problems they would be celebrated (as much as mathematicians are).They solved the problem, that's all there is. Bravo.
If they didn't cite correctly, fix that. Nothing wrong with fixing a paper letter. -
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."
@timnitGebru At least the headline places blame correctly: OpenAI committed research misconduct. A person engages in conduct, one conducts oneself, and their their tools (the LLMs they used) execute operations but don't have a self to conduct.
