aldo booze

Gross little bugs, squish em

[aldo booze @Ravenoil]

@TonyPannWBAL @wbaltv11 Thank you mister weather man.

[aldo booze @Ravenoil]

@llallawg a big part of AI’s current success is its ability to throw a ton of (reasonably but not extremely deep) strategies at a ton of problems way faster than a human can: this would likely involve *tons* of computing power spent on problems it ultimately couldn’t solve

[aldo booze @Ravenoil]

@llallawg Very possible (and probably very likely), but it is a big difference at the current moment. Especially since the current AI resolutions to major problems, while impressive, are not as “deep” as you would expect given the nature of the problem. Suggests to me that (1/2)

[aldo booze @Ravenoil]

@_NathanCalvin from a problem of such importance. In this case, the proof relies heavily on much deeper human results that get you a lot of the way there. To me what the AI is best at right now is trying *a lot* of reasonable (but not super deep) proof strategies, far faster than humans can

[aldo booze @Ravenoil]

@_NathanCalvin It’s true that this is one of the more important problems in GGT, though I would not put it in the top 50 most important problems in math (there is a lot of math out there!) interestingly, similar to the Jacobian one, the proof is a lot “easier” than you’d expect (1/2)

[aldo booze @Ravenoil]

@theojaffee “Several of these would have individually been the result of the decade” — is this talking about the 10 results Astra produced? These are very major results, but that statement doesn’t seem true at all.

[aldo booze @Ravenoil]

@Rohit_Writes Fair enough, maybe this wasn’t the best post to express this sentiment (I’ve heard of many of these problems as well). I’ve just heard “my chatgpt instance disproved a 30 year old conjecture” one too many times lol. This is undoubtedly huge news though

[aldo booze @Ravenoil]

@Rohit_Writes (this is especially important when people say they disproved a "30-year old graph theory conjecture" and the conjecture itself is just one of a list of 200 or whatever that were themselves generated by a computer)

[aldo booze @Ravenoil]

@Rohit_Writes In general I want to say (as someone in math academia) that how long ago a conjecture was posed isn't really too indicative of how difficult or important it is. I might be biased but the sofic groups result seems to be far and away the most major, but it wasn't posed too long ago

[aldo booze @Ravenoil]

@emollick Even as a math person (phd student), I feel this a ton about results outside of my expertise. the sofic groups result has been the first that's in my field! (My take: it's very impressive and also relies heavily on much deeper human work, so I don't feel useless quite yet haha)

[aldo booze @Ravenoil]

@littmath First large AI result I've seen in geometric group theory (my "domain"). It is very cool, and at least for now I'll content myself with the fact that GPT's example eseems to rely heavily on much deeper prior (human) results in the same direction. Might change soon, though...

[aldo booze @Ravenoil]

@Usama7Hassa @0xThoughtVector @SebastienBubeck No, it's an important example and a big deal but (1) I wouldn't consider it fields medal worthy on its own and (2) the work done in the actual paper is very heavily built on a few prior (human) papers that seem a good bit deeper than what the AI did.

[aldo booze @Ravenoil]

@seanluomdphd @stalkermustang @EpochAIResearch This is not true. It would perhaps be mentioned in a (graduate-level) geometric group theory textbook, though the word "sofic" does not appear in the two that I checked. Still an important result in geometric group theory.

[aldo booze @Ravenoil]

@Carbon19992025 @captgouda24 While this is often true, there’s nothing wrong or really objectionable about this post

[aldo booze @Ravenoil]

@bjorkienne @LinkofSunshine @HotOompaLoompa I find it very believable because (1) the "brute force" search space for this problem is far too large (I guess unless you smartly restrict it), and (2) the proof given is quite natural, and is pretty simple! to me that kind of proof is totally in reach of current models

[aldo booze @Ravenoil]

@tracewoodgrains Like, what this reinforces for me is just that there’s so much more mathematics to do than there are people doing it!

[aldo booze @Ravenoil]

@tracewoodgrains Incoming PhD student with slim-to-moderate arithmetic geometry knowledge here, I get why Litt reacted that way. *I* could understand the argument, and it feels very natural… how was this unresolved for this long?? With some guiding hints this could be a textbook problem!

[aldo booze @Ravenoil]

@lthlnkso This guy also isn’t a normal math PhD. To me he feels like the type of guy to solve this conjecture in a world without AI models

[aldo booze @Ravenoil]

@antifataylorjoy i think likely a bit higher (even if you remove all of its televote points it still places 17th overall), but probably something like 13th-15th. definitely none of their past 3 songs would be top 5 contenders

[aldo booze @Ravenoil]

@kriisifrutti nkahdeksan