QotNews Hacker News, Reddit, Lobsters, and Tildes articles rendered in reader mode.
Sam Altman sympathizes with mathematicians, but says OpenAI’s math onslaught won’t stop. Altman says OpenAI has spent hundreds of millions on math — and has more discoveries up its sleeve.
I'm a mathematics professor at research university.
The community is more concerned on how to reorganize itself. We are not afraid that Mathematics will be over (impossible, doesn't even make sense) or that we will lose our jobs (possible in some future, but that would require academia to collapse).
I’m hearing concern that people don’t want to
Study anymore. Concern that the publishing is impossible now, some journals paused accepting any new papers
Because people are writing so many papers using LLMs that they have way more submissions than ever before, more than can be read using the old methods. I think this will shake out to be a reactionary response, but this is sort of tearing down the walls of academic publishing! Open AI is releasing things on GitHub, not even on arxiv! It’s a strange new era for academia
Spreading everywhere that „AI solved math“ has plenty of adverse effects.
Plenty of young people will likely decide against going into mathematics. Non-mathematicians deciding who gets how much funding will give math departments less money. Etc.
The main question is how to assign value to work. We've been judging mathematicians basically by how impressive the proofs they've been able to find have been. That metric is now completely broken because AI can generate far more proofs than the community is able to process. That means we now have to assign credit to something else, but what's that supposed to be? Formulating the most "interesting" questions? Achieving a certain level of understanding of AI proofs? Those things are extremely difficult to measure. Nobody is quite sure what to do right now.
>Altman says OpenAI has spent hundreds of millions on math
Which is on the same order of magnitude as the entire federal budget for basic math and stats research. The question then is if that's actually buying us more than what we were already doing - we know that research grants are translating into thousands of peer-reviewed papers every year, but what we don't know is OpenAi's results stack up - the bluster around them ignores the fact that we literally haven't had time to fully verify them, or that they have yet to meet the same standards any run of the mill mathematician is held to. It's a big question mark at this point.
This needs to get repeated everywhere. How much more productive would all this money have been just giving the resources these companies have used to actual mathematicians?
Imagine instead if we spent that $10 million on actual research programs for human mathematicians. Thousands of scholars could have received research assistantships with that money. Instead it's all been spent to call these big problems "solved" despite no human being any closer to understanding it, and no other benefit achieved other than OpenAI declaring victory on their self-assigned computational pissing contest.
Human mathematicians already solved shitloads of problems and got stuck with problems which are hard for humans to solve. Throwing more money on humans to solve those problems wouldn't help much.
Makes more sense to try to solve those problems with AI and have humans try to verify those solutions.
Solving those problems is not just about giving mathematicians something to have fun with. These solutions result in advancements in technology.
Why would we do that? Research shouldn't be a jobs program. Solving these problems absolutely advances the state of mathematical knowledge, even if what most mathematicians now do is use LLMs to understand these proofs.
Right, but this missed the point. This is a young and quickly improving technology. The question isn't a static comparison between AI and month ago and old-style basic research at the moment. It's more interesting to consider what returns we get a year from now, or five years from now, in terms of research gains, by investing money in improving AI today.
Right, so do we agree or what? Because you said "the question then is if that's actually buying us more than we were already doing", and I'm saying, on one interpretation of that sentence, that that is not the question.
If you want to compare the research returns five years from now, you still have to compare investing that money in AI against investing it in conventional mathematical research over the same period.
You're not correcting the question. You're changing it.
That comparison only gets worse for human mathematicians, no? Human research is not scalable in this way. Doubling the research money afforded to mathematics doesn’t guarantee an increase in output, there are a fixed number of mathematicians that operate with fixed maximum productivities. And the time lag to train more is enormous.
Given equal time the effects become exaggerated in favor of AI seemingly.
Reviewing complicated proofs can take months, sometimes years. When the Poincare conjecture was solved it took almost a decade for it to be widely accepted enough to get the millennium prize.
Were they? Have you verified any of the proofs? Has anybody verified all of them? You can have a lean proof verification compile and not throw any 'sorry' results that does not verify the exact thing you research claims it does
The difference stops here: mathematicians don't get any smarter or cheaper as the months go by. A Navier–Stokes-level problem will cost pennies in five years. It's a fight you can't win as a human.
The other thing is to see what the limit of capability is. Thing about compute and token cost is that it now cost 1/100th of what it was not too long ago. One of the fastest deflation in price of a technology we’ve seen. So many crazy things going on now.
They've certainly done more progress in math than what a year of federal grants has. If the federal government knew they could just invest more and solve one millennium problem, make significant progress on four more, as well as a plethora of other advances, they would not have a problem investing many times as much.
>They've certainly done more progress in math than what a year of federal grants has. If the federal government knew they could just invest more and solve one millennium problem
According to the standards set out for a Millennium prize, it literally isn't certain.
Have you seen the response from the Clay Institute? They were pretty excited about it. Their process takes a minimum of two years so of course it's not given immediately.
To be fair, the hundreds of millions is mostly on training, and even then, it would have taken much more for humans to have done what they did. If you assume that the average openai problem would be solved by a human in 2 years, then assuming $200k yearly salary and if they've solved 1k problems, thats 200k*2*1000=400 million.
You’ve got private money going into math matching the public money. Purely on economics a doubling is a huge win. It’s almost idle debate if private money is getting the returns of public money. In no way am I led to believe pub funds by way of grants will stop funding traditional math research before this question is resolved
I'm not sure that I agree that AI isn't progressing maths understanding. The proofs themselves might not be doing much compared to if teams of human beings had worked to produce them, but there's a lot of evidence that maths researchers are widely using AI to understand and perform research in a way that's facilitating their productivity — and not just in the superficial, slop-oriented sense. You even see this documented in reports from mathematicians like Terrence Tao. I think it's a mistake to treat AI's impact on mathematics as only coming from proof generation.
Or out right stolen from someone else’s work before they finished the project. It’s not that simple and it is at the same time. The reason we pursue these math problems is for the journey not the end goal. We want to learn new math, new proofs, new ways of solving complex problems that we are able to create new frame works around. Same reason why we do research in everything, we don’t know where it may lead us. Most recent example is a peptide in the Gila monster leading to glp1s
Edit: the way ai solves the problems isn’t very cohesive or follows along how a human would think and rationalize. It brute forces the answers to these problems. I am not anti ai because I do see the value in it as a tool to help us accomplish things. But theres a difference in having the ai do it for you vs the ai helping you. An example would be how a lot of people are using ai to do their research and write their arguments/papers for them and when you actually want to discuss things they arent actually able to explain why or how with more depth and understanding.
Can you please explain how can you “steal” from a math book? What does it even mean? You know that stuff written in the math books is just mostly theorems and their proofs?
Technically, mathematics is not a field of science. It does not deal with any observations or experimentation. It has many scientific applications, and mathematic discoveries could perhaps one day inform or assist in a scientific discovery. But it is not a science in and of itself.
I can see why OpenAI keeps claiming otherwise though. They want to be able to say their models have made scientific discoveries. But it's simply not true.
Even putting that aside, many mathematicians have been pointing out these so called "discoveries" are practically useless to the field.
>Both follow a very similar process of observing areas yet to be explained
I mean, if we want to get pedantic, mathematics is a priori. There may be similarities in how mathematical and scientific questions are explored, but there's a fundamental epistemological distinction between proving a mathematical proposition and empirically testing a scientific hypothesis.
And I will say that for a long period, math kind of followed from what we were observing about the real world. The distinction is that the actual "rules" put in place for mathematics aren't contingent on empirical observations. Once you establish a set of axioms and definitions, you can derive conclusions independently of whether they correspond to anything in the physical world.
There is. Mathematics is a formal system; something AI is quite good at dealing with. Very distinct in terms of how its advanced to scientific discovery, which is not a formal system, and so operates by very different principles.
It's not a requirement of being a mathematician to know that mathematics is not a field of science.
Mathematics is a formal system. Science, while it often utilises formal systems, is itself not a formal system.
Formal systems do not rely on experimentation or observations of reality; they have axioms, rules and then they have theorems that are either valid or invalid. Very different goals and methods to science.
Mathematics is not a formal system. You are so full of shit, also “not to be a formal system” is not really a criteria for something to be a science. Also, theorems are not either valid of invalid.
Also, math absolutely rely on observations of reality. Because mathematical facts are absolutely part of our reality.
Maybe this would've been more clear if before today you asked yourself where and what you are inferencing, because science is built upon inferences. The places mathematics is used alongside inferencing are called mathemetical sciences, for example statistics
the fact that I we this comment verbatim any time an article like this pops up will be looked upon poorly in the history books. Strikes me as poisoning the well.
The fact that we have formulas for distributions that model the usage of human language and mathematics to the extent that they can fill in the cracks in our knowledge and understanding of the world by producing novel discoveries is amazing to me. Our societies aren’t adjusted to this new possibility yet but that doesn’t mean that we shouldn’t explore them. Opportunity necessitates change.
There is nothing stopping you, from going backwards and figuring out how the answer was derived.
If alien technology crash landed on Earth, we would go bananas trying to reverse engineer it. The same is true with these results. The work is now reverse engineering which is much more productive than wasting time on less productive approaches.
That is now how academia is set up, and, at least for now, no one is offering to pay people to do that. Highly educated people aren't going to pivot to doing free work for a multi-trillion dollar industry just because. I'm sure many mathematicians would do it if someone paid them to, but for now, Anthropic expects it to be done for them with no further investment.
This is a good application of AI - scientific advancement. Sorry for people who’s salary is attached to the a “mathematician” title, but they’ll need to learn to use AI at their advantage, just like software engineers need to learn to use Claude and other tools at their advantage.
Not a single mathematician lost their job to AI and it won’t be happening in the foreseeable future.
If anything AI generated more work for mathematicians as there are know hundreds of messy indecipherable proofs that are only useful if someone understands them.
Cute_Presence_2861 | 4 hours ago
I'm a mathematics professor at research university.
The community is more concerned on how to reorganize itself. We are not afraid that Mathematics will be over (impossible, doesn't even make sense) or that we will lose our jobs (possible in some future, but that would require academia to collapse).
NaVa9 | 4 hours ago
What's the main concern then when it comes to reorganizing itself? In what ways?
doiwantacookie | 4 hours ago
I’m hearing concern that people don’t want to
Study anymore. Concern that the publishing is impossible now, some journals paused accepting any new papers
Lopsided_Tiger_0296 | 3 hours ago
Publishing articles is mostly a scam anyways as experiments are designed and data manipulated for the results the scientists want
Trick_Pin_2139 | 3 hours ago
not in mathematics it ain't 💀
slimeystevebob | an hour ago
Mathematics is not an empirical science 🤦
NaVa9 | 3 hours ago
Source?
Spare-Dingo-531 | 3 hours ago
Why don't journals want to accept new papers?
Maybe it's a good thing Openai is dumping all the progress they are making on GitHub, if journals don't want to publish papers.
doiwantacookie | 3 hours ago
Because people are writing so many papers using LLMs that they have way more submissions than ever before, more than can be read using the old methods. I think this will shake out to be a reactionary response, but this is sort of tearing down the walls of academic publishing! Open AI is releasing things on GitHub, not even on arxiv! It’s a strange new era for academia
_ECMO_ | 3 hours ago
Spreading everywhere that „AI solved math“ has plenty of adverse effects.
Plenty of young people will likely decide against going into mathematics. Non-mathematicians deciding who gets how much funding will give math departments less money. Etc.
aasfourasfar | an hour ago
Maths or any formal system is fundamentally unsolvable because it has to contain true statements which have no proof. See Gödel incompleteness
_ECMO_ | an hour ago
I am aware. Most people who keep hearing the propaganda are not.
aasfourasfar | 43 minutes ago
Holy smokes .. I think I replied to the wrong comment
MediumMachineGun | an hour ago
Yeah but the next level blow that is "solving everything that is solvable", which only leaves the unsolvable.
TomvdZ | 46 minutes ago
The main question is how to assign value to work. We've been judging mathematicians basically by how impressive the proofs they've been able to find have been. That metric is now completely broken because AI can generate far more proofs than the community is able to process. That means we now have to assign credit to something else, but what's that supposed to be? Formulating the most "interesting" questions? Achieving a certain level of understanding of AI proofs? Those things are extremely difficult to measure. Nobody is quite sure what to do right now.
Disastrous_Room_927 | 5 hours ago
>Altman says OpenAI has spent hundreds of millions on math
Which is on the same order of magnitude as the entire federal budget for basic math and stats research. The question then is if that's actually buying us more than what we were already doing - we know that research grants are translating into thousands of peer-reviewed papers every year, but what we don't know is OpenAi's results stack up - the bluster around them ignores the fact that we literally haven't had time to fully verify them, or that they have yet to meet the same standards any run of the mill mathematician is held to. It's a big question mark at this point.
asyork | 4 hours ago
And the expectation is that the mathematicians will verify it for OpenAI without being paid by OpenAI, while OpenAI tries to make them irrelevant.
Disastrous_Room_927 | 4 hours ago
Convenient that they're releasing things that are logistically challenging to verify.
BagOfWetToiletPaper | 3 hours ago
This needs to get repeated everywhere. How much more productive would all this money have been just giving the resources these companies have used to actual mathematicians?
Necessary_Job3578 | 46 minutes ago
Do you think OpenAI doesn’t have mathematicians on their team?
They literally have a few field medalists on their full time pay roll.
AI assisted math is the future. We have to accept it and move on.
strange_username58 | 4 minutes ago
We were no where near solving the millennium problems no matter how much money was thrown at marh.
LordSprinkleman | an hour ago
And how much more productive will AI be not just now, but a few months from now? A year from now?
Are you intentionally missing the much more important big picture here?
snozzd | 3 hours ago
Imagine instead if we spent that $10 million on actual research programs for human mathematicians. Thousands of scholars could have received research assistantships with that money. Instead it's all been spent to call these big problems "solved" despite no human being any closer to understanding it, and no other benefit achieved other than OpenAI declaring victory on their self-assigned computational pissing contest.
PunkPirate56364 | 18 minutes ago
Human mathematicians already solved shitloads of problems and got stuck with problems which are hard for humans to solve. Throwing more money on humans to solve those problems wouldn't help much.
Makes more sense to try to solve those problems with AI and have humans try to verify those solutions.
Solving those problems is not just about giving mathematicians something to have fun with. These solutions result in advancements in technology.
TWAndrewz | 42 minutes ago
Why would we do that? Research shouldn't be a jobs program. Solving these problems absolutely advances the state of mathematical knowledge, even if what most mathematicians now do is use LLMs to understand these proofs.
Famous-Football-4617 | an hour ago
I don't want to wait centuries to get my results.
Murky-Hold-2901 | 3 hours ago
Right, but this missed the point. This is a young and quickly improving technology. The question isn't a static comparison between AI and month ago and old-style basic research at the moment. It's more interesting to consider what returns we get a year from now, or five years from now, in terms of research gains, by investing money in improving AI today.
Disastrous_Room_927 | 3 hours ago
>Right, but this missed the point. This is a young and quickly improving technology.
I'm one of the people whose profession is improving this technology. I feel like it'd be hard for me to miss the point here.
Murky-Hold-2901 | 3 hours ago
Right, so do we agree or what? Because you said "the question then is if that's actually buying us more than we were already doing", and I'm saying, on one interpretation of that sentence, that that is not the question.
Disastrous_Room_927 | 3 hours ago
If you want to compare the research returns five years from now, you still have to compare investing that money in AI against investing it in conventional mathematical research over the same period.
You're not correcting the question. You're changing it.
garret1033 | 2 hours ago
That comparison only gets worse for human mathematicians, no? Human research is not scalable in this way. Doubling the research money afforded to mathematics doesn’t guarantee an increase in output, there are a fixed number of mathematicians that operate with fixed maximum productivities. And the time lag to train more is enormous.
Given equal time the effects become exaggerated in favor of AI seemingly.
Murky-Hold-2901 | 2 hours ago
Yeah, that's correct. But that is the more important comparison to make.
Spare-Dingo-531 | 3 hours ago
What specifically is your problem? Many of the results are already verified in lean and now we can go and reverse engineer them.
shumpitostick | 3 hours ago
People act like it's never been a problem before.
Reviewing complicated proofs can take months, sometimes years. When the Poincare conjecture was solved it took almost a decade for it to be widely accepted enough to get the millennium prize.
Infamous_Mud482 | an hour ago
Were they? Have you verified any of the proofs? Has anybody verified all of them? You can have a lean proof verification compile and not throw any 'sorry' results that does not verify the exact thing you research claims it does
Mediocre_Gap2236 | 10 minutes ago
Terrance Tao has looked at a couple
Is he not good enough for your standards
Many_Consequence_337 | an hour ago
The difference stops here: mathematicians don't get any smarter or cheaper as the months go by. A Navier–Stokes-level problem will cost pennies in five years. It's a fight you can't win as a human.
Busterlimes | 3 hours ago
No its not, its mumbojumbo. If we dont understand how AI got the answer, we cant verify it. Let OpenAI piss away billions.
ensui67 | an hour ago
The other thing is to see what the limit of capability is. Thing about compute and token cost is that it now cost 1/100th of what it was not too long ago. One of the fastest deflation in price of a technology we’ve seen. So many crazy things going on now.
phido3000 | 4 hours ago
Yeh but this is paid in debt and destroy humanity.
shumpitostick | 3 hours ago
They've certainly done more progress in math than what a year of federal grants has. If the federal government knew they could just invest more and solve one millennium problem, make significant progress on four more, as well as a plethora of other advances, they would not have a problem investing many times as much.
Disastrous_Room_927 | 3 hours ago
>They've certainly done more progress in math than what a year of federal grants has. If the federal government knew they could just invest more and solve one millennium problem
According to the standards set out for a Millennium prize, it literally isn't certain.
shumpitostick | 2 hours ago
Have you seen the response from the Clay Institute? They were pretty excited about it. Their process takes a minimum of two years so of course it's not given immediately.
Recent-Peanut6061 | 3 hours ago
To be fair, the hundreds of millions is mostly on training, and even then, it would have taken much more for humans to have done what they did. If you assume that the average openai problem would be solved by a human in 2 years, then assuming $200k yearly salary and if they've solved 1k problems, thats 200k*2*1000=400 million.
TyrellCo | 3 hours ago
You’ve got private money going into math matching the public money. Purely on economics a doubling is a huge win. It’s almost idle debate if private money is getting the returns of public money. In no way am I led to believe pub funds by way of grants will stop funding traditional math research before this question is resolved
justneurostuff | 2 hours ago
I mean, even if it's not buying more per dollar, it's still apparently a doubling of spending on basic math and stats research. Yay?
Thormidable | an hour ago
But proving it isn't usually particularly useful. The maths understanding we develop to prove it is. But AI isn't progressing that
justneurostuff | an hour ago
I'm not sure that I agree that AI isn't progressing maths understanding. The proofs themselves might not be doing much compared to if teams of human beings had worked to produce them, but there's a lot of evidence that maths researchers are widely using AI to understand and perform research in a way that's facilitating their productivity — and not just in the superficial, slop-oriented sense. You even see this documented in reports from mathematicians like Terrence Tao. I think it's a mistake to treat AI's impact on mathematics as only coming from proof generation.
Brahm-Etc | 4 hours ago
"Spent hundreds of millions on math" or stole it from books to "train" their models?
hustl3tree5 | 4 hours ago
Or out right stolen from someone else’s work before they finished the project. It’s not that simple and it is at the same time. The reason we pursue these math problems is for the journey not the end goal. We want to learn new math, new proofs, new ways of solving complex problems that we are able to create new frame works around. Same reason why we do research in everything, we don’t know where it may lead us. Most recent example is a peptide in the Gila monster leading to glp1s
Edit: the way ai solves the problems isn’t very cohesive or follows along how a human would think and rationalize. It brute forces the answers to these problems. I am not anti ai because I do see the value in it as a tool to help us accomplish things. But theres a difference in having the ai do it for you vs the ai helping you. An example would be how a lot of people are using ai to do their research and write their arguments/papers for them and when you actually want to discuss things they arent actually able to explain why or how with more depth and understanding.
Glum-Objective3328 | 4 hours ago
That will all continue to happen. People will read the proofs, and ask new questions.
Spare-Dingo-531 | 3 hours ago
It literally could not have being stolen from books. Almost all of the proofs are entirely original results.
The exponential time hypothesis disproof in particular was highly unexpected.
aaron_moon_dev | an hour ago
Can you please explain how can you “steal” from a math book? What does it even mean? You know that stuff written in the math books is just mostly theorems and their proofs?
Iorith | 4 hours ago
Just like students "steal from books" to train their brains.
asking_porn | an hour ago
At least they pay for the books
paranoid_throwaway51 | an hour ago
Typically, companies don't sell students.
Sufficient-Skill9530 | 4 hours ago
I can’t wait for his company to shit the bed and we can all forget about this creepy alleged sister rapist
Spare-Dingo-531 | 3 hours ago
OpenAI will be bought by Microsoft before it disappears.
And we should be grateful to people like Sam Altman for bringing AI into the world.
JudgeMyReinhold | 5 hours ago
They've spent hundreds of millions scouring wolfram and stack overflow? Cool
Glum-Objective3328 | 3 hours ago
Was Navier Stokes, Quasi Reinman, Jacobian Conjecture on Wolfram or stack overflow?
Spare-Dingo-531 | 3 hours ago
What a dismissive and foolish result then groundbreaking science!
The quasi rieman hypothesis, for example, is absolutely enormous. And openai made one of the biggest contributions to solving it in decades.
SunbornSamson | an hour ago
Go ahead and scour wolfram and try to solve millenium problems open for decades
New_Salamander_4592 | 3 hours ago
are those discoveries also some other mathematicians unreleased work that got fed into their matrix math machine?
SunbornSamson | an hour ago
Who knew the biggest obstecle to math would be mathematicians
Waste-Falcon2185 | an hour ago
Literally why? It's not like they earn (much) money from this? Just to inflate the IPO?
A_Poor_Miser | 3 hours ago
I'm tired of hearing about this monster.
Decentralizator | 11 minutes ago
Sometimes I wonder if they wish for all of us to die.
Hungry_Prior940 | an hour ago
I hope it continues.
Dry-Competition8492 | an hour ago
OpenAI wont make it until 2029
Responsible-Laugh590 | 5 hours ago
Anyone trying to stop scientific discoveries will be looked upon poorly in the history books
CognitionMass | 5 hours ago
Technically, mathematics is not a field of science. It does not deal with any observations or experimentation. It has many scientific applications, and mathematic discoveries could perhaps one day inform or assist in a scientific discovery. But it is not a science in and of itself.
I can see why OpenAI keeps claiming otherwise though. They want to be able to say their models have made scientific discoveries. But it's simply not true.
Even putting that aside, many mathematicians have been pointing out these so called "discoveries" are practically useless to the field.
Eye_foran_Eye | 4 hours ago
AI stole a mathematicians work and then did the last 2% and said "Look what I did!"
Recent-Peanut6061 | 3 hours ago
You clearly have no knowledge about mathematics
Iorith | 4 hours ago
Yes just like every mathematician has done off the backs of their professors.
Spare-Dingo-531 | 3 hours ago
That's a lie. That is not true.
Kuriente | 4 hours ago
>mathematics is not a field of science
While technically true, I'm not sure there's an important enough distinction between the two to place a hard barrier.
Both follow a very similar process of observing areas yet to be explained, pursuing explanations, and testing explanations.
Disastrous_Room_927 | 4 hours ago
>Both follow a very similar process of observing areas yet to be explained
I mean, if we want to get pedantic, mathematics is a priori. There may be similarities in how mathematical and scientific questions are explored, but there's a fundamental epistemological distinction between proving a mathematical proposition and empirically testing a scientific hypothesis.
And I will say that for a long period, math kind of followed from what we were observing about the real world. The distinction is that the actual "rules" put in place for mathematics aren't contingent on empirical observations. Once you establish a set of axioms and definitions, you can derive conclusions independently of whether they correspond to anything in the physical world.
CognitionMass | 2 hours ago
There is. Mathematics is a formal system; something AI is quite good at dealing with. Very distinct in terms of how its advanced to scientific discovery, which is not a formal system, and so operates by very different principles.
Cute_Presence_2861 | 4 hours ago
Mathematician for 10 years here. Today I learned that mathematics is not a science.
CognitionMass | 3 hours ago
It's not a requirement of being a mathematician to know that mathematics is not a field of science.
Mathematics is a formal system. Science, while it often utilises formal systems, is itself not a formal system.
Formal systems do not rely on experimentation or observations of reality; they have axioms, rules and then they have theorems that are either valid or invalid. Very different goals and methods to science.
aaron_moon_dev | an hour ago
Mathematics is not a formal system. You are so full of shit, also “not to be a formal system” is not really a criteria for something to be a science. Also, theorems are not either valid of invalid.
Also, math absolutely rely on observations of reality. Because mathematical facts are absolutely part of our reality.
Infamous_Mud482 | an hour ago
Maybe this would've been more clear if before today you asked yourself where and what you are inferencing, because science is built upon inferences. The places mathematics is used alongside inferencing are called mathemetical sciences, for example statistics
Glum-Objective3328 | 3 hours ago
Feels like a semantical response
Disastrous_Room_927 | 4 hours ago
the fact that I we this comment verbatim any time an article like this pops up will be looked upon poorly in the history books. Strikes me as poisoning the well.
Scared_Weird6087 | 5 hours ago
Good
Spare-Dingo-531 | 3 hours ago
Agree. Keep the results coming!
ballsosteele | an hour ago
He's really, really, really desperate to impress those investors isn't he?
Difficult-Nobody-453 | an hour ago
ai psychotic
skinnyjoints | 5 hours ago
The fact that we have formulas for distributions that model the usage of human language and mathematics to the extent that they can fill in the cracks in our knowledge and understanding of the world by producing novel discoveries is amazing to me. Our societies aren’t adjusted to this new possibility yet but that doesn’t mean that we shouldn’t explore them. Opportunity necessitates change.
_ECMO_ | 3 hours ago
AI math proofs are novel discovery in the sense that Douglas Adams‘ „42“ was a novel discovery.
Useless in real life.
asyork | 4 hours ago
> fill in the cracks in our knowledge and understanding
It is generally viewed as the exact opposite of that when you are given an answer without first figuring out how to do it yourself.
Spare-Dingo-531 | 3 hours ago
There is nothing stopping you, from going backwards and figuring out how the answer was derived.
If alien technology crash landed on Earth, we would go bananas trying to reverse engineer it. The same is true with these results. The work is now reverse engineering which is much more productive than wasting time on less productive approaches.
asyork | an hour ago
That is now how academia is set up, and, at least for now, no one is offering to pay people to do that. Highly educated people aren't going to pivot to doing free work for a multi-trillion dollar industry just because. I'm sure many mathematicians would do it if someone paid them to, but for now, Anthropic expects it to be done for them with no further investment.
lowkeytokay | 4 hours ago
This is a good application of AI - scientific advancement. Sorry for people who’s salary is attached to the a “mathematician” title, but they’ll need to learn to use AI at their advantage, just like software engineers need to learn to use Claude and other tools at their advantage.
_ECMO_ | 3 hours ago
Not a single mathematician lost their job to AI and it won’t be happening in the foreseeable future.
If anything AI generated more work for mathematicians as there are know hundreds of messy indecipherable proofs that are only useful if someone understands them.
Disastrous_Room_927 | 4 hours ago
Bad bot
lowkeytokay | 4 hours ago
YOU are the bad bot
Disastrous_Room_927 | 4 hours ago
I know you are but what am I?
lowkeytokay | 4 hours ago
You are a disaster… a disastrous room, to be precise
Disastrous_Room_927 | 3 hours ago
let's see how long we can keep this going: that's your mom's favorite room.
lowkeytokay | 2 hours ago
Well… she has bad taste
Spare-Dingo-531 | 3 hours ago
Bad bot.