Hacker Newsnew | past | comments | ask | show | jobs | submit | acedTrex's commentslogin

> second place

If the result holds then this is the paper that will get the medal and what history will remember. First place enough in my book.


ChatGPT already found issues with it, as per this comment chain. Argon also spots a bunch of errors

Looks like he already knows (and is mad about that too).

It has to be new people that have come here, this place was NEVER this terrible before. Now it seems like the world has conspired to direct the worlds greatest idiots to descend on this poor site.

> the latest LLMs have a very low false positive rate for bugs.

Man people live in different worlds, astra for me finds 25/6 bugs all the time.


This is easily the funniest thing i've read in some time. All possible motivations for typing it are hilarious.

I'm confused. I just meant to say that some people have given the paper to astra and astra has said it's wrong. That's all.

It's a very "pack it up boys, we're done here" comment and the context makes it hilarious. Like the guy who spent 12 years writing his work should "Oh right, my bad" and go on doing something else.

The aggressive uselessness of the comment makes it funny.


If you look at my post history, I'm a mathematician. I've fully conceded, painfully, heartwrenchingly, that Astra is better at math than I'll ever be.

I wish it wasn't the case but it's true.


That does not make your comment any more valuable though

Your original comment is both of little to no substance AND highly ironic, given the current state of affairs regarding this.

It contains no actual counter proof, does not say which people have done so, does not say what the models said, does not say which models and does not even give your own opinion or a unique point of view.

It is little more than "Some people have done something and say nuh-uh"


ok, here's the alleged mistake in the proof

mistake in the proof: https://philpapers.org/archive/NIEIOM.pdf


This is only a subset of alleged errors, and not the problematic ones.

i agree with your first statement, not sure about your second (was this identified as a reparable error by any LLMs?)

> People have fed it to an LLM

> I'm a mathematician

And hearsay is how you do maths?


Sorry, but LLMs are better than anyone alive at math at this point. I will take what they say at face value most of the time, especially if it's a subject I have no expertise in. I know it's an appeal to authority but if you can find me a math problem an LLM cannot figure out and a human can, I will happily reconsider. Right now those don't exist.

> if you can find me a math problem an LLM cannot figure out and a human can, I will happily reconsider. Right now those don't exist.

Yes they do. I'm a mathematician (formerly depressed PhD student, graduated into merely depressed) and your outlook is too bleak. Consider, did you get into math because you like it, or because it was socially prestigious and you were better than your peers at it?

Ask an LLM to solve an involved origami crease pattern (say "Old Dragon" by Travis Nolan), with a step by step fold sequence. It will fail. Amateur human origamists can do it, and it's a well-defined mathematical problem.


I'm curious for some elaboration on the proposed problem - are you suggesting that "Take a paper with dimensions x, fold as follows, then describe the result" is something an LLM couldn't currently answer correctly?

My initial feeling is that this is something an LLM would be quite capable of, and the gap is likely a tooling and specification issue (i.e. you should first develop a method for simulating origami folds, then build on that)

A non-savant human certainly couldn't answer the question as posed, either. They would answer it by actually following the steps and seeing what happened.

Have I misunderstood what you're proposing?


Look up "Old Dragon" by Travis Nolan. I chose it because it's an intermediate-level model with, as far as I can tell, no published step-by-step fold sequence or video tutorial. There is only the (monochromatic) crease pattern, i.e. what the model looks like when completely unfolded. I am asking for a fold sequence given the crease pattern.

Given that description, it might sound about as hard as reassembling a shredded document. But the truth is that it doesn't take a savant when you're familiar with the most common patterns, and that's the reason why folders online generally share their models this way. Yet legit academic research on doing this with computers even for simple models is primitive (see Creasy/ORIZU for the closest prior attempt, or more recently OrigamiBench documenting essentially "LLMs suck at this, even with tooling"; if you dig around you'll find some interesting work by Lang suggesting that the admittedly quite general technique of tree theory that he has algorithmically pioneered since the 90s can be driven by AI, but that falls far short of a general understanding of origami).

Anyway, obvious caveats being: I am under no delusion that AI couldn't become superhuman at this given a clever enough vectorization of the problem amenable to deep learning, and/or a huge corpus of richly labeled training data, but I stand by my claim this is a well-defined mathematical problem type that current AI cannot solve just by asking it, while a skilled human can.


Thanks, that's a detailed and interesting response to think about. It suggests that humans have some heuristic that efficiently collapses the huge tree of possible fold sequences into a practically testable set of candidates. I guess that's obvious, but it's interesting to think about what gives us that ability.

Indeed, these problems have been considered plenty already, and even though it's an NP-hard minefield, the patterns humans actually create and read are much better behaved.

Some further clarifications on the exact problem:

- You can assume you're given either the final ordering as input data (as it won't be unique to the given crease pattern, even with mountain-valley assignments), or that you can target _any_ valid final layer order; - One has to settle on a model for what folds are physically plausible. It's quite easy to formalize a model for the most common and locally tame folds in origami, but arbitrary collapses and things like reverse sinks may require a fully 3D and time dynamical model of origami, such as a souped up spring-hinge model with elasticity and collision handling.

AI has not been very helpful for me in exploring the latter. It demands its hand be held as it cannot judge the fidelity–tractability tradeoff for itself at all.


I love doing math, but it has an opportunity cost. Sitting and thinking about a problem for a long time feels less and less attractive these days, especially when I could just go outside and have fun or do other things with my life. Math doesn't need me anymore. I might as well be Calvin in the Calvin and Hobbes comics at this point.

You said you would happily reconsider in light of an example, but you didn't comment on it. Have you reconsidered or haven't you?

> I love doing math, but it has an opportunity cost.

I'm not sure you love doing math if you think of it as an opportunity cost rather than as an opportunity. You love what it earns you, not math itself.

Anyway, I'm in the same boat. I finished my PhD already having come to the conclusion that I, too, did not have as big of an appetite for math as I thought (I don't regret pushing myself through the burnout at the expense of some mental health, either). I still indulge math and research from time to time, but with this lack of attachment to results or social approval, I find AI invigorating rather than despairing.

Anyway, I am also very glad I finished my studies before AI took off. I don't envy your position. Please take care and be lenient with yourself; you're not the only one wondering WTF is going on or what we're supposed to be doing right now.


"I love doing math"

Maybe you just love going outside and doing other things more than you love math and it's not really about AI at all.


You might not like it, but that's just how math and programming work in 2026, and it's the right approach. The gap will just increase.

[flagged]


Wasn't trying to and am not claiming to have made a riveting contribution.

When everyone can "build" something, nothing "built" matters. It was fun while it lasted, blessed I got to enjoy a niche for as long as I did.

I get the sentiment, but there's something really nice now about picking a project and just chipping away at it for months until you have something you really like. That was a lot harder outside of work int he past, and now I feel like I can take notes when I have ideas and set an agent to work on it later. I've been learning a lot, and really enjoy thinking about software a step up from the line and module level thinking that used to take up most of my time.

Alpoge did not build the forklift, kind of a basic flaw in that analogy.

Don't think Claude comes with a 'Solve historical math hypothesis' button.

Pretty sure he didn't just prompt 'Solve Riemann, make no mistakes'


I can not fathom this viewpoint, it is as alien to me as being a flat earther. What frame of mind is required to consider an analogy of "we built saturn V instead of walking" to be even remotely ontologically in the same stratosphere as what is happening today?

I can't even begin to make sense of that lack of basic understanding of humanity.


It turns out that the cognitive processes of advanced maths are subject to mechanization, just like carrying a load from point A to point B. It is a blow to our self notion that work of the mind is somehow fundamentally distinct from work of the rest of the body.

Also, people are nervous that the mechanization of new human capabilities will give even more power to money and less to labor, and we will live in some distopian libertarian society. I think that is a separate and realistic fear from the grief of discovering our thoughts are not as special as we told ourselves.

For me, if I am confident that the social structure will enable fulfilling lives for people, my kids and their peers, then I am ok with exploring and understanding the possibilities of mechanized cognition.

And to those objecting to the phrase mechanized cognition, I use Claude a lot and the thing I find missing is that independence of perspective that all the people I encounter have. We wake up with a perspective and goals and preferences, a why to answer what built into the protoplasm. Some technology will rebuild that aspect of our experience in a different substrate, but it doesn’t feel like a system that can’t count but can prove Riemann is quite wired correctly for that fleshy self motivated loop.


Giving a shit is decimated in this day and age, everyone just slops shit and breaks things.

In a way companies are forcibly lowering the bar on the "give-a-shit" side to demonstrate productivity gains from AI.

They were already doing that before AI, just an endless queue of tickets you're told what to do, how to do, and when they have to be done by. You can't give a meaningful shit within those confines and nobody wants you to.

It reminds me of a Michael Scott quote from The Office -

> Don't ever, for any reason, do anything to anyone for any reason ever, no matter what, no matter where, or who, or who you are with, or where you are going, or where you've been... ever, for any reason whatsoever...

Just get those tickets done or they'll find someone who will.


Ridiculous field we are in, people seemingly have forgotten the absolute basics in under a year.

The people doing these things never knew the basics.

i think the main deal is speed right?

No, the main deal is that it CANNOT answer with anything outside the given structured options. LLMs can go off the rails, however small a chance it is. The architecture behind JEV and the like doesn't let it do that.

> structured json outputs

Right. Producing JSON still requires the decode part of inference, which is not as compute efficient. Decider models are in some sense equivalent to producing only a single token. At least in performance terms.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: