Announcement on X: link.
(and if you wanna see some of the usual suspects on HN doing their best to cope, here you go)
Summary:
- The conjecture in question is the cycle double cover conjecture, and resolving it has been a major open problem in graph theory for ~50 years.
The statement of the conjecture is actually pretty simple and immediately understandable to CPers:
For any bridge-less undirected graph $$$G(V, E)$$$, you can find a set of cycles $$$C$$$ such that each edge in $$$E$$$ is covered by exactly two cycles in $$$C$$$.
- Shockingly, the proof is also not that illegible for a mathematically illiterate person like me, and I was able to get a high-level understanding after talking to AI about it for an hour.
- Unlike previous instances of AI labs solving open math problems, this proof was produced entirely by a publicly available model (5.6 Sol Ultra, running 64 agents in parallel in codex).
- The cost for this particular attempt at proving this particular problem likely wasn't exorbitant at all (we can estimate the cost quite easily as we know that it involved 64 agents running collectively for under an hour, which translates to just a few hundred dollars based on API prices (due to an upper-bound on the output's tok/s)). This obviously excludes the cost of all the failed attempts they would have internally made at resolving other open problems with the same harness, but that's not really relevant imo.
- The proof has been formalized in Lean here.
It seems reasonable to expect most humans to be obsolete by 2028. I wonder what it feels like to be someone entering a math/cs undergraduate program right now.
Do you think the next millennium problem to be solved will be solved autonomously by AI? and if so, when?
Lets see what happens ... Like we can't even do anything at this point :>
Wow!
llms are good at solving nt/comb math problems but still struggle with some areas of math. And as far as I know, llms are not that good at solving kaggle problems. You never know how long it will take for llm based models to overcome their weaknesses.
I remember seeing an image of tung tung sahur on X, but it was labelled "cope cope sahur" instead. If this comment were posted by a >= blue, I would have considered wasting time looking for that image and then posting it here. As things stand, it's not worth it.
its good that you didnt post that picture because its not related to my comment. If you r struggling to understand my comment, ask gpt to explain it to you.
You are also in the same boat as us in terms of being obsolete. Their capabilities have surpassed even those of the best coders here, let alone a master like you.
I don't know when LLMs will make every human obsolete, but they already respond more constructively than many humans.
Said the person who did 8 rounds for 2 years. How about getting off your high horse and participating before looking down on others?
If you're using graytist, how were you able to read this comment?
by clicking the "show" button
You don't have to be wonder how it feels to be entering Maths/CS undergraduate program right now, it extremely depressing, and everyday imagining myself being unemployed even after Graduation!
CP is keeping me alive ig!
I don't know, it seems like there is just so much more to most jobs than whatever the current $$$LLMs$$$ can do. Especially for jobs where you have to get up and do something in real life. But yeah, I think $$$LLMs$$$ can definitely take away a lot (not all) of tech jobs, but then again a lot of those jobs didn't exist $$$20$$$ years ago. So it's probably just gonna be one of those things where new technology is introduced and is the final nail in the coffin for a bunch of transient jobs, and everyone continues on more efficiently because of the new technology.
Of course what $$$LLMs$$$ can do rn wrt math and stuff is extremely impressive, but I really highly doubt that $$$LLMs$$$ will be the cause of mass job loss.
Consider any job that can be done remotely (ie. almost entirely through a computer interface), which is a criterion almost all tech jobs satisfy. Do you think there's any fundamental barrier that will prevent AI from being able to do such work better, faster, and cheaper than humans by eoy-2028? If so, can you name some concrete examples?
This argument gets repeated a lot, but AI just isn't comparable to any technology of the past. In the limit, I don't see how humans retain any relative advantage over AI at stuff that can be done entirely through a computer interface.
There are many tech jobs that require physical interaction with systems, like safety-critical embedded systems, high performance computing, and hardware security. Not all jobs can be done "remotely through an interface"
I believe that the short timeline is the biggest barrier here. The end of $$$2028$$$ is $$$2.5$$$ years from now, and I really don't see $$$AI$$$ being able to do entry-level tech support better, cheaper, faster than most workers in that timeframe. And even if it could, it's a fairly big project for companies to actually fire all their workers and implement some $$$AI$$$ system to do all of that, so there would still be a lot of tech support roles out there.
Also, my experience so far of companies trying to entirely automate jobs has been disappointing. Like they can't even automate the ordering process at taco bell yet. I was there the other day and they had removed the $$$AI$$$ robot they put in place since it would mess so many things up.
I've been thinking about it recently, and I really believe that the magnitude of change that has been and will be brought about by $$$LLMs$$$ (and maybe even $$$AI$$$ in general) is comparable to those of past developments. Honestly, I'm still a little shocked that you can paste an incredibly hard cf problem into a modern $$$LLM$$$ and it will generate a perfect solution usually in less than a minute. But imagine how much it shocked people of the past when they could instantly talk to each other from thousands of miles away, or travel across an entire continent in the air in just a few hours. I really do think that developments like the nuclear bomb, the fact that people gained the ability to destroy entire countries in minutes, will have caused more change than $$$LLMs$$$. And I bet there are some older people today who still have trouble accepting the fact that you can store many millions of encyclopedias of information in the palm of your hand.
But none of those changes really caused any sort of cataclysm (except in Japan). In fact, people have benefitted and I think have become even more useful, on average, due to (most of) these developments. $$$AI$$$ will change a lot of things, definitely, but I don't think that they will render most people useless.
I agree with what you said but I have also been worried about the quality of jobs that remain in the AI era. The most interesting part of the job can be done using AI leaving the tedious boring parts for the humans. I see this even now in software engineering where most of my core work and brain storming are done via a prompt and I am just a verifier with very little of my own inputs.
Great, another AI doomer post. Thanks DNR! I found about this tool recently that blocks posts with a specific keyword, I'm going to use that.
greatest comeback ever
if those entering a math/cs undergraduate program right now can't get an ideal job,than there's nobody in the world can.
how do you justify this statement ? its highly likely that software jobs will be automated before (and to a higher degree), than for example, doctors, lawyers, politicians or plumbers
What matters, I guess, is those that still involve people: dialect, human-computer interaction, accessibility for older folks......
Update (20/07/2026): Claude Fable 5 has resolved the Jacobian Conjecture (reference).
I don't even know what to think anymore :( Why did AI have to come up in the exact time I'd be getting a job after a degree T_T It's depressing as hell
The prompt says: "Do not search the public web merely to determine whether CDC is open, and do not answer that it is open."
The prompt writer knows the problem is open. The prompt writer is trying to force the AI to pretend it solved it. The prompt writer is role-playing a breakthrough. This is not serious research. This is a game.