> Regardless of the true cost, it seems that professional mathematicians now need to wary about what they put into a LLM and think hard about how to disclose and publish a result.
This is all but guaranteed now.
Mathematicians/Scientists/Researchers need to stop sharing freely with "AI Companies" and have explicit clauses in place in their publications about not using their research without their explicit consent.
There should be a clear legal distinction between using research data for AI model-training vs. another researcher using it.
Come up with a legal framework, establish procedures for sharing and using others work and have a single scientific body in charge of enforcing it.
Just putting a clause in a publication won't prevent it from being used as training data. Information wants to be free.
The frontier LLM vendors do sell enterprise licenses which contractually guarantee that your prompts won't be used for training. (Maybe they'll secretly violate the agreement but in principle it's legally enforceable.) Scholars and universities who care about credit and attribution will either have to purchase those licenses or run their own private open-weight LLM instances.
Even the $20 tier of ChatGPT has privacy settings that forbid using the user's data to be used for training. The question is, whether this setting is respected.
minus two or three things, corporate data is classified as public/goverment data. we just saw something about earmarking domain last month? two being imminent domain. three being natsec.
risk of prescient theory is more important than dismissive ablation.
edit: to wit, facebook google and anything else not e2e.
Would this legal framework cut both ways? When AI companies use AI to make and publish mathematical discoveries, would they be able to legally prevent professional mathematicians from using them?
This is about cooperation before publishing results. And they will keep everything medieval secret, else some big company steals it and claims it their own.
So then the solution is to publish more often (e.g. on a public blog) even if your ideas are not fully developed in order to establish priority and show you are doing something.
Analogously with software development, it's always been good practice to write things down, but since the start of this year it's become dramatically more important for everyday work.
Will they do that? Obviously they want some publicity about their model capabilities recently, but even there they have their eyes on major newsworthy problems. Are they going to scan everyone's blogs so they can scoop them on some random paper on e.g. PDEs modeling plasma dynamics in different gasses? Why? They just want to spite researchers and are willing to burn money doing it?
> Despite the never-ending thirst for knowledge of all branches of the order, most Tech-priests of the Adeptus Mechanicus have lost the ability to innovate or carry out basic scientific research.
> No longer the master of its creations, the Cult Mechanicus is enslaved to the past. It maintains the glories of yesteryear with rite, dogma and edict instead of true discernment and comprehension.
> For instance, even the theoretically simple process of activating a vehicle's engine is preceded by the application of ritual oils, the burning of sacred resins and the chanting of long and complex hymns.
What if those chants are prompts? "Make no mistakes", "My Grandma used to love this", "fix this and don't break anything else". Sometimes you have to keep chanting for a while until you get the results.
There are sources claiming AI is Abominable Intelligence and something completely different from the Machine Spirit. But it seems to me that this is how it feels to be a tech priest.
Stop treating AI companies and their software as somehow unconstrained, above-the-law actors. It's delusional that anyone buys that. Regulate them appropriately.
At the same time, mathematicians should be using sophisticated, specialized LLM tools in much more sophisticated ways than lay people. There should be no way lay people can compete. There are new tools to master and if you use your slide rule, you won't keep up. It's a chance for mathematics productivity to boom.
With apologies to Baudelaire: The greatest trick exploitative forces ever played was convincing the people that it was impossible to imagine anything else.
If in 2019 someone wrote a program that generated photorealistic child porn on demand, the FBI would break down their door the day after hearing about it.
Like we can prevent the rest of our society from devolving into the Medieval Era of secrecy: by treating individuals with respect and dignity and not as the ore from which resources can be profitably extracted.
Prior to LLMs there was some minimal effort required to snipe someone and possibly your reputation was attached otherwise there would have been no point in publishing to begin with.
Now anyone with a few dollars can do it, many who don't have a reputation to worry about.
It's the same problem as YouTube AI slop, AI-generated music, and everything else. Don't you dare tweet or blog about a video idea or hum a few bars from a song - within an hour 27 people will have posted AI slop rip-offs. There's something uniquely depressing about being beaten to the punch by a thief that doesn't deliver the same soul-crushing impact as having someone copy you after the fact.
Definitly seeing this with video games. Especially ones that are easy to make and have generic names. I mean you could already easily clone those before, but now 10 people will easily clone them.
I think making these type of games simply becomes an unviable source of income in the current era.
I'm sure there's a named concept for this: the idea of the game is its selling point, and has to be revealed to consumers to market itself. But the market has low friction (via Steam distribution, and by copies being cheap to make w/ LLMs), so you can't reliably establish your product before masses take over, especially if you don't have reputation built up from previous games.
You have to be bringing in something else for this not to happen, trade secrets not present to the consumer being one such factor.
You could patent your game design, if that's possible. This is exactly what patents were created for. But you yourself are probably churning out original (instead of copycat) little games and can't afford to patent all of them.
I don't belive you can patent gameplay. You can't say "I'm going to patent the idea that you all go around a board buying land and getting cash, if your player moves to someone else's land they have to pay them rent."
But you can copyright the artistic aspects of your game, like you can have the copyright to MONOPOLY and trademark the logo.
If you could patent gameplay there would be only 1 pong, which is a pretty basic idea. There are probably hundreds of pongs though.
It's a screenshot of a list of games on Steam. Half of them are called "Needle in a Haystack" and the rest are variations. The title says the game went viral a week ago and the original creator had to put "(original)" in the title.
I wonder how many fields will have to suffer through their collective work being sloppily copied or used as a basis for generative slop before we figure out we aren't getting AGI just Artificial General Slop.
I welcome the era of secrecy. After living so much in this era of open information where everyone seems to know everything, secrets may be a way to make things more interesting again.
There is a financial incentive to selling access to the leading LLMs needed to find whatever secret result there is out there. See the play station hypervisor 0day from the other day. If a LLM can find someone's secret 0day they are flaunting around they can find a math proof someone else says they have.
1. People will publish AI generated frontier math making it difficult to identify frontier mathematicians.
2. Trained frontier mathematicians will become scarce
99.99% of finance is freshman level math if at that. Then 99.99% of the remaining 0.01% is somewhat more complicated but still not frontier level.
There are very-very few cases where you need truly advanced math in finance. The problems that need solving are generally vastly more pedestrian.
You need to deal with terrible data quality, terrible formats, noise in every aspect of your work, disruptions, lack of standards, inability to generate new data (and repeat experiments), conflicting and often opaque incentives, technical problems ranging from shitty APIs to having to squeeze nanoseconds out of your network stack, etc.
These are all difficult problems, but they are crucially not frontier math problems (by and large).
> How to admit you're in finance without admitting you're in finance?
Tell me you have no clue about quant finance without telling me you have no clue about quant finance.
You probably think it’s abstract topology and Ito calculus, when in reality it’s linear regression and PCA. If you’re lucky, maybe you’ll see a sigmoid function.
If someone in finance is using LLMs, it’s for marketing purposes (recruiting students, or impressing investors). Jane Street/HRT/DE Shaw are notorious for this.
> This is mostly true, but there are exceptions (e.g. Renaissance Technologies).
Nope.
Let me guess: you read the is The Man Who Solved the Market and think they use/used undisclosed frontier mathematics to predict the market.
There are plenty of firms with both higher annual % returns (consistently over 15-20 years) and higher absolute $ returns than RenTech/Medallion fund, especially post-2010.
A lot of “alpha” comes from how well you know the VP at the exchange. Not math. Not tech.
Don’t like a market participant? Tell the exchange to issue that pesky participant market violations. Get them banned for a few months.
Fat fingered a trade with wrong price or quantity? The exchange can undo a trade after the fact. All in the name of “orderly markets”. Just like how governments use “national security” as a blanket excuse.
It's been in use far longer than LLMs. Thought I'm not exactly sure when it came into common use. I'm pretty sure I've heard the term on Nova decades ago.
Yea, for sure AI brought it into the mainstream. The terms existed for a long time, but the number of people actually involved in it are really low so it's not going to show up much on trends unless something big happens in the field.
(edit: this is books data)ngrams says first use in ~1889 in use for a few years, flatline until re-introduction in ~'66, surprisingly peaking in ~'83, and current sitting at roughly less than half of peak useage.
Though it appears to have gained popularity from 2017-2021, it dipped in '22 and that's as far as ngram goes.
That's also what the gerrymandering discourse is like. Gerrymandering is a threat to civilization because it means political parties will minimize their electoral margins, and because it means politicians will maximize their electoral margins.
The whole point of gerrymandering is to split districts so that votes for the other party are diluted. The whole point is to create many districts where your party is majority and then a few districts for the rest of the opposition.
If you are good at it, end result is that minority can keep majority of the seats and power. So, as there are two parties, the groups are "likely to voted republicans" and "likely to vote democrats".
There are many conjectures that we know are almost surely true, but we don't know why, and explaining why is the main purpose of the mathematician when publishing a proof.
The fact that we don't know why is a clue pointing at some area of math that we haven't discovered yet. The hope is always that it will uncover some hidden fertile valley that will lead to lots of new discoveries. But the proof of the conjecture itself, without understanding, is really not that valuable.
My point is that even if AI discovers many new truths, there's still plenty to do for the mathematical community, in dissecting it and building useful abstractions to understand it, abstractions that can be leveraged for further exploration and uncovering new questions.
Seems perfectly possible to get the worst of both worlds: more mathematics than anyone can read, and less access to the mathematics people actually care about.
> 1. People will publish so much frontier mathematics, humans won't be able to understand it all
That already happened before AI.
> 2. Frontier mathematics will all be kept secret
Gauss kept lots of frontier mathematics in his drawer. In the 20th centuries government spy agencies developed public key cryptography long before that was known to the public. To give just two examples.
It's not the end of the world.
And what do you care, if someone keeps frontier mathematics a secret, if you can ask DeepSeek version 10 in 2030 to prove the Riemann hypothesis for you?
About 4-6 years earlier, depending what you want to count, although the inventors may also have been less clear on its importance or applications compared to the later public inventors.
I do think they can happen together like malnutrition and obesity. Pursuit of recognition, intellectual property, remuneration, war machines, etc, will naturally drive new strategies for academics to compete with capital-intensive AI research. One strategy will surely be new types of silo-ing, whether that will kill the post-enlightenment spirit of math we’ve come to love is still an open question
The secrecy will be volumetric. Im sure it will continue to be possible for bullshit to out-generate information. With bullshit being redefined to true, but informationally/conceptually useless proofs.
Universities should be providing university hosted llms to their faculty and students. No student or faculty member should be using public llms for their work. They can share info directly with other people or in non-public forums to keep it away from commercial llms. It is going to have to be against the rules to submit anyone else's work to a commercial llm. Businesses are going to have similar policies.
Universities can host the llms just like they hosted any other computer lab or web service on campus. This is what universities are supposed to be doing. Universities should be involved in open model research and should offer models that are not datamined.
Or if you are afraid of scooping, just don't use public LLMs? At universities, apart from admin stuff, nobody will tell you what you must or musn't use.
And I really doubt university ITs would be able to run these, at least in an up to date and stable manner. High performance computing is typically run totally separately from everyday user systems, and tend to be extremely painful to use (hello slurm!).
It's today's computer lab or server time. This is something colleges would have always provided in house. It is very strange that colleges aren't doing what they normally would do.
The problem here is that AI isn't just a problem, it also revealed a problem: too much emphasis on publishing papers and churning out new results. The name of the game right now is just that: make something new and significant. AI itelf is a problem around the world right now because we've set up some seriously bad incentives. Same with art - it devolved into content creation for money, so AI snaps that up.
What we need to do is make mathematics about understanding, rather than churning out results. People should be rewarded for reaching an ability to explain mathematics without the aid of computers, to teach people for the sake of their learning.
I do think AI also needs to be eradicated because of its destructive properties, but I think that at the same time it also is a manifestation of the sickness in our society to go after the wrong incentives that are detrimental in the long run.
I wonder how life was while the Butlerian Jihad was raging through the universe.
Jihadis were fighting opponents with thinking machines. Did they win because the machines were not capable enough?
The idea that this genie is going back in the bottle is, imo, very wishful thinking. The thinking machines can rip off your software ideas completely, in days.
This. The value of intelligence and execution is plummeting towards zero for knowledge work. Right now, software and mathematics are the most affected, but the transition to every other industry is going to happen with shocking speed.
We routinely close the feedback loop when training people in pretty much every industry. Otherwise, what would you say that exams and accreditation/chartering are? I don't see anything inherent in other knowledge work that means AI couldn't do it.
I respect your opinion, and I was with you all the way until your last sentence.
"Eradicated" is a pretty strong word. It's also completely unfeasible.
Wrong incentives can maybe be adjusted for. Shutting down the pursuit of one of the most astonishing things we've ever created is simply not going to happen short of a cataclysm, and I'm not a big fan of those.
True, eradicated is a strong word and I do think it should be. And as horrible as a cataclysm is, I think it's still going to be less pain for humanity than going forward with this tech.
This is a simple matter of intellectual property. A tool or technique for doing a thing can be patented. A legal monopoly is granted to the originator. The public is made aware of the technique but is legally forbidden from using it for a period of years/decades.
Talking about frontier mathematics? Even the common for-mere-mortals version of mathematics is totally inaccessible to most people. Even to those who want to really learn it.
Take a good look at the textbooks that teach mathematics and tell me I am wrong.
I am well aware of all the sites and YT, what you probably thought I live in a cave? I taught mathematics at Tsinghua for over a decade, so I know what I'm talking about.
3B1Br is great but only for a 'hobbyist'. There's no systematic progression to teach from A to Z. Also, you can't learn mathematics just by watching videos.
Whether intentional or not, it's there. I have yet to see a textbook that really taught learners as they deserve to be taught. Some are good in some aspects, but a perfect textbook does not exist.
"Here are some theorems and some proofs and here are some ideas left as exercise to the student". smh.
I think "final" here just means the unit that currently gets treated as publishable and credit-worthy: a theorem with a proof, a solved conjecture, a completed paper, etc
If it's not worth publishing, it's not a result at all. That being said, the problem really is that since LLMs excel at sifting through piles or papers, publishing more would aggravate the problem instead of alleviating it.
All of life proceeds effectively with a right brain and a left brain. A fast loop and a slow loop. A general and a scout. A melody and a base. The evolutionary moments involve moving the melody into the base and making way for a new melody. But there is always an overseeing element and a work element. A manager or coach and an employee or athlete. It's just an effective pattern of growth. You see it everywhere in nature. In generations of animals and plant life (the more experienced parent, the growing child; the central 'brain' and the rest of the body; the queen ant and the ant colony; the trunk and the leaves).
The friction in math right now is you have the fast moving melodic bits racing faster than the base can understand always or keep up with. So you start to need AI for both the left brain part that is executing and the right brain part that is synthesizing. That's just the friction right now. You reduce the friction by using AI to help with understanding AI and getting back to a more normal rhythm of scouting out (with AI) and (what is still developing more and more) synthesizing with AI.
Make the proof from LLMs with the same constraints as human mathematicians have:
The proof must be as succinct as possible so that it's reviewable. It's ok for the same LLM to publish repeatedly, but it's not Ok to drop a bulk 600kb Lean proof.
The proof must build on other works that are similarly bite sized and published before (so that it's incrementally reviewed)
The proof must be published as soon as fully formed, and not in bulk like current LLMs proofs.
This will not happen. The fear comes from the real problem: publication as a measure.
The mathematics itself will only benefit from the discovery of cross links from the human created literature. Think about these LLM as infinitely patient and very long attention experts in what was already done.
So on one side, for a few years at least, it will become normal to publish tens of articles. The article inflation is what scares the present system.
On the other side, after a while, when all the existing mathematical corpus will be mined and most of buried connections will be explored, we shall have a much more better foundation for future mathematics.
It is very naive or misleading to think about mathematics as if it were chess or go.
In few year these (future) tools will be in standard use.
127 comments
[ 0.24 ms ] story [ 64.5 ms ] thread> Maybe mathematicians are smart enough to never ever buy such piece of shit on higher principle
They are not. Mathematicians are as human as most of the rest of us here.
This is all but guaranteed now.
Mathematicians/Scientists/Researchers need to stop sharing freely with "AI Companies" and have explicit clauses in place in their publications about not using their research without their explicit consent.
There should be a clear legal distinction between using research data for AI model-training vs. another researcher using it.
Come up with a legal framework, establish procedures for sharing and using others work and have a single scientific body in charge of enforcing it.
The frontier LLM vendors do sell enterprise licenses which contractually guarantee that your prompts won't be used for training. (Maybe they'll secretly violate the agreement but in principle it's legally enforceable.) Scholars and universities who care about credit and attribution will either have to purchase those licenses or run their own private open-weight LLM instances.
risk of prescient theory is more important than dismissive ablation.
edit: to wit, facebook google and anything else not e2e.
it's not like the ai is homomorphic.
"government intelligence agencies" ARE 'AI'.
Though maybe if there's a flurry of math-optimized agents coming up, like there are small coding agents, those might be feasible to host personally.
Analogously with software development, it's always been good practice to write things down, but since the start of this year it's become dramatically more important for everyday work.
Just, without the aliens, space travel, or mech-suits.
Got half the other downsides though!
Your joke overestimates the intelligence of real life humans.
https://warhammer40k.fandom.com/wiki/Tech-Priest:
> Despite the never-ending thirst for knowledge of all branches of the order, most Tech-priests of the Adeptus Mechanicus have lost the ability to innovate or carry out basic scientific research.
> No longer the master of its creations, the Cult Mechanicus is enslaved to the past. It maintains the glories of yesteryear with rite, dogma and edict instead of true discernment and comprehension.
> For instance, even the theoretically simple process of activating a vehicle's engine is preceded by the application of ritual oils, the burning of sacred resins and the chanting of long and complex hymns.
There are sources claiming AI is Abominable Intelligence and something completely different from the Machine Spirit. But it seems to me that this is how it feels to be a tech priest.
At the same time, mathematicians should be using sophisticated, specialized LLM tools in much more sophisticated ways than lay people. There should be no way lay people can compete. There are new tools to master and if you use your slide rule, you won't keep up. It's a chance for mathematics productivity to boom.
With apologies to Baudelaire: The greatest trick exploitative forces ever played was convincing the people that it was impossible to imagine anything else.
Prior to LLMs there was some minimal effort required to snipe someone and possibly your reputation was attached otherwise there would have been no point in publishing to begin with.
Now anyone with a few dollars can do it, many who don't have a reputation to worry about.
It's the same problem as YouTube AI slop, AI-generated music, and everything else. Don't you dare tweet or blog about a video idea or hum a few bars from a song - within an hour 27 people will have posted AI slop rip-offs. There's something uniquely depressing about being beaten to the punch by a thief that doesn't deliver the same soul-crushing impact as having someone copy you after the fact.
See this amusing reddit post. https://www.reddit.com/r/IndieDev/comments/1wfl64l/game_went...
I'm sure there's a named concept for this: the idea of the game is its selling point, and has to be revealed to consumers to market itself. But the market has low friction (via Steam distribution, and by copies being cheap to make w/ LLMs), so you can't reliably establish your product before masses take over, especially if you don't have reputation built up from previous games.
You have to be bringing in something else for this not to happen, trade secrets not present to the consumer being one such factor.
But you can copyright the artistic aspects of your game, like you can have the copyright to MONOPOLY and trademark the logo.
If you could patent gameplay there would be only 1 pong, which is a pretty basic idea. There are probably hundreds of pongs though.
It's a screenshot of a list of games on Steam. Half of them are called "Needle in a Haystack" and the rest are variations. The title says the game went viral a week ago and the original creator had to put "(original)" in the title.
1. People will publish so much frontier mathematics, humans won't be able to understand it all
2. Frontier mathematics will all be kept secret
Fortunately, these seem like they can't both happen at once.
1. People will publish AI generated frontier math making it difficult to identify frontier mathematicians. 2. Trained frontier mathematicians will become scarce
" The National Security Agency is the largest employer of mathematicians in the U.S. "
https://www.scientificamerican.com/article/mathematicians-an...
There are very-very few cases where you need truly advanced math in finance. The problems that need solving are generally vastly more pedestrian.
You need to deal with terrible data quality, terrible formats, noise in every aspect of your work, disruptions, lack of standards, inability to generate new data (and repeat experiments), conflicting and often opaque incentives, technical problems ranging from shitty APIs to having to squeeze nanoseconds out of your network stack, etc.
These are all difficult problems, but they are crucially not frontier math problems (by and large).
Tell me you have no clue about quant finance without telling me you have no clue about quant finance.
You probably think it’s abstract topology and Ito calculus, when in reality it’s linear regression and PCA. If you’re lucky, maybe you’ll see a sigmoid function.
If someone in finance is using LLMs, it’s for marketing purposes (recruiting students, or impressing investors). Jane Street/HRT/DE Shaw are notorious for this.
Nope.
Let me guess: you read the is The Man Who Solved the Market and think they use/used undisclosed frontier mathematics to predict the market.
There are plenty of firms with both higher annual % returns (consistently over 15-20 years) and higher absolute $ returns than RenTech/Medallion fund, especially post-2010.
A lot of “alpha” comes from how well you know the VP at the exchange. Not math. Not tech.
Don’t like a market participant? Tell the exchange to issue that pesky participant market violations. Get them banned for a few months.
Fat fingered a trade with wrong price or quantity? The exchange can undo a trade after the fact. All in the name of “orderly markets”. Just like how governments use “national security” as a blanket excuse.
No, more like, some differential geometry and gauge theory. The topics that Jim Simons worked on.
Though it appears to have gained popularity from 2017-2021, it dipped in '22 and that's as far as ngram goes.
Which makes total sense unlike the "too much frontier math" nonsense.
Which groups are those?
If you are good at it, end result is that minority can keep majority of the seats and power. So, as there are two parties, the groups are "likely to voted republicans" and "likely to vote democrats".
According to you, there is no discourse about politicians using gerrymandering to give themselves safer seats.
The fact that we don't know why is a clue pointing at some area of math that we haven't discovered yet. The hope is always that it will uncover some hidden fertile valley that will lead to lots of new discoveries. But the proof of the conjecture itself, without understanding, is really not that valuable.
My point is that even if AI discovers many new truths, there's still plenty to do for the mathematical community, in dissecting it and building useful abstractions to understand it, abstractions that can be leveraged for further exploration and uncovering new questions.
That already happened before AI.
> 2. Frontier mathematics will all be kept secret
Gauss kept lots of frontier mathematics in his drawer. In the 20th centuries government spy agencies developed public key cryptography long before that was known to the public. To give just two examples.
It's not the end of the world.
And what do you care, if someone keeps frontier mathematics a secret, if you can ask DeepSeek version 10 in 2030 to prove the Riemann hypothesis for you?
About 4-6 years earlier, depending what you want to count, although the inventors may also have been less clear on its importance or applications compared to the later public inventors.
https://en.wikipedia.org/wiki/Public-key_cryptography#Classi...
I guess that's kind of a long time in computer technology terms.
Universities can host the llms just like they hosted any other computer lab or web service on campus. This is what universities are supposed to be doing. Universities should be involved in open model research and should offer models that are not datamined.
And I really doubt university ITs would be able to run these, at least in an up to date and stable manner. High performance computing is typically run totally separately from everyday user systems, and tend to be extremely painful to use (hello slurm!).
What we need to do is make mathematics about understanding, rather than churning out results. People should be rewarded for reaching an ability to explain mathematics without the aid of computers, to teach people for the sake of their learning.
I do think AI also needs to be eradicated because of its destructive properties, but I think that at the same time it also is a manifestation of the sickness in our society to go after the wrong incentives that are detrimental in the long run.
I wonder how life was while the Butlerian Jihad was raging through the universe.
Jihadis were fighting opponents with thinking machines. Did they win because the machines were not capable enough?
The idea that this genie is going back in the bottle is, imo, very wishful thinking. The thinking machines can rip off your software ideas completely, in days.
Automated spreadsheets and PPTs won't get people fired.
"Eradicated" is a pretty strong word. It's also completely unfeasible.
Wrong incentives can maybe be adjusted for. Shutting down the pursuit of one of the most astonishing things we've ever created is simply not going to happen short of a cataclysm, and I'm not a big fan of those.
Take a good look at the textbooks that teach mathematics and tell me I am wrong.
3B1Br is great but only for a 'hobbyist'. There's no systematic progression to teach from A to Z. Also, you can't learn mathematics just by watching videos.
Part of the difficulty is artificial.
Whether intentional or not, it's there. I have yet to see a textbook that really taught learners as they deserve to be taught. Some are good in some aspects, but a perfect textbook does not exist.
"Here are some theorems and some proofs and here are some ideas left as exercise to the student". smh.
Maybe we should start crediting and publishing our intermediate results and attempts.
Humans would gain credit for providing a part of the solution to tough problems, LLMs would profit from the resulting data flywheel.
> Maybe we should start crediting and publishing our intermediate results and attempts.
All mathematical results are "intermediate results". What would it mean for something to be a final result?
The friction in math right now is you have the fast moving melodic bits racing faster than the base can understand always or keep up with. So you start to need AI for both the left brain part that is executing and the right brain part that is synthesizing. That's just the friction right now. You reduce the friction by using AI to help with understanding AI and getting back to a more normal rhythm of scouting out (with AI) and (what is still developing more and more) synthesizing with AI.
The simple short-term solution is just don't feed all your work into the IP theft machine?
The proof must be as succinct as possible so that it's reviewable. It's ok for the same LLM to publish repeatedly, but it's not Ok to drop a bulk 600kb Lean proof.
The proof must build on other works that are similarly bite sized and published before (so that it's incrementally reviewed)
The proof must be published as soon as fully formed, and not in bulk like current LLMs proofs.
The mathematics itself will only benefit from the discovery of cross links from the human created literature. Think about these LLM as infinitely patient and very long attention experts in what was already done.
So on one side, for a few years at least, it will become normal to publish tens of articles. The article inflation is what scares the present system.
On the other side, after a while, when all the existing mathematical corpus will be mined and most of buried connections will be explored, we shall have a much more better foundation for future mathematics.
It is very naive or misleading to think about mathematics as if it were chess or go.
In few year these (future) tools will be in standard use.