It does not. If after careful analysis people conclude that they cannot tell if it says what it claims to say, then then it can be considered not-a-proof until someone can demonstrate in a convincing manner that it is what it claims to be.
I can say god exists because 5>3. No matter how valid my claim is that five is in-fact greater than three, if I can't convince people of the connection it's a valid proof of something, just not what I say it is.
It might be true that if the Riemann hypothesis is false, leaves cannot fall to the ground. That would mean a falling leaf demonstrates the truth of the hypothesis. Without showing people why it must be the case in a way they understand it falls short of a genuine proof.
> Airliners are still subsonic after a 100 years of aviation.
Yeah, but that's a choice made for, arguably, very good reasons:
- Noise reduction
- Fuel efficiency
- Reduced pollution and lower environment impact
- Reduced cost of materials, equipment, and maintenance
- Passenger affordability
It's not that we can't make supersonic airliners: it's that it doesn't make economic or environmental sense to.
Gulfstream and Sukhoi even had a partnership to develop a supersonic business jet (which, if you look at the mockups, might be one of the coolest looking planes ever conceived) but they gave up after spending a fairly enormous amount of money already because it was going to cost an absolute fortune to bring to market... and there was no real market for it.
I can argue all of the reasons are valid for keeping compute at a certain speed range, or power consumption envelope.
It's not we can't make compute much more denser, there's a new standard informally called as megawatt racks. But deploying them at gigantic scales bring in massive problems as well, so we can keep total energy envelope at a fixed range.
> Airliners are still subsonic after a 100 years of aviation
That is a deliberate decision motivated primarily by economics and secondarily by environmental concerns. In fact airliners today are somewhat slower than they were a few decades ago because the lower speeds are more fuel efficient. We certainly could build fleets of supersonic airliners if it made sense to do so, it is not as if we have no idea how to do it and in fact we built supersonic passenger jets once upon a time.
Your example contradicts your point. Airliners are still subsonic after an extended period of time, and hence based on that past performance it's a good indicator (although not definitive) that they will remain subsonic for an extended period into the future.
In general, past performance is actually indicative of future results. It's not definitive, and some phenomenon are random, but the trajectory of technology over time is not some random fluke that just has coincidentally trended by random chance.
>Your example contradicts your point. Airliners are still subsonic after an extended period of time, and hence based on that past performance it's a good indicator (although not definitive) that they will remain subsonic for an extended period into the future.
There was a period in aviation development when it was widely assumed that speed will continue to rise. Just like in the development of CPUs there was such a period, quite recently.
But along other axes they have also improved tremendously, no? Ie rather than just time to destination, they have greatly improved efficiency (energy spent per unit of work) and throughput/parallelism (ie simultaneous flights in the air) right?
Yes, but these axes are also not showing an unlimited growth. There is no expectation that either speed, fuel efficiency, ticket price, or total number of flights will improve by an order of magnitude in the foreseeable future, unless there's a technological breakthrough (Nuclear turbines? Artificial photosynthesis making kerosene free? 3D printing the entire plane? These are currently all sci-fi ideas).
Sure you're going to get a lot of math problems solved, but food distribution is still going to be stuck by warlords and oligarchies.
Think of it like highway designers and the "one more lane" gridlock delusion. In this case, it's more will smith eating speghetti than cheap, low cost cancer cures.
Rather, nobody wants them at the price point necessary to make them economically viable. In the context of tech, the cost to develop a next-gen fab doubles every four years; absent some major breakthrough, we have about 20 years before even the most advanced economies balk at spending $1T, $2T, $4T, $8T on a single fab, causing hardware to settle down like airline speeds have.
the best computing power used to be located at the universities and national laboratories. Now you need access - employee or otherwise - to a nasdaq listed startup. Hard not to think it's the shareholders who benefit in the short term in this new alignment.
What an offputting title. The sneer really comes through. If I ever had any desire to read what this man wrote, this little potshot has extinguished it.
It's good to poke fun at the AI phrase that's aged the worst and all the news articles that promoted it as fact. It did a lot of damage to people's understanding of AI.
Not off-putting in the slightest. It's an well-deserved jab at the earliest LLM naysayers whose "criticism" was just ignorance wrapped in causeless activism. They claimed that LLMs were just randomly repeating locally coherent phrases back to us and that no amount of scaling could produce real intelligence. Well, the goal posts have moved far indeed if a proof of GRH isn't intelligence.
> My 9-year-old son, informed that RH was proved, replied that he’s not even surprised at this point. Maybe you need to have come of age in the human math era to be surprised by such things.
His 9-year-old son is more intelligent than you are. I'm not even surprised at this point.
Scott Aaronson has more intellectual chops and a better track record than 99.999% of people commenting on AI news. The sneer is a small fraction of the sneer coming from the know-nothings who have been using phrases like "stochastic parrot", "fancy autocomplete", etc. for many years now. The parrots are coming home to roost, as they say.
Seriously. We are less prepared for the present and future because of all the denialists and naysayers, many on HN, who have repeatedly cried "stochastic parrot", "it just regurgitates known things, it can't create" and "marketing stunt" endlessly, sometimes >80% of the sentiment on threads over the past years.
Just a proof of the infinite monkey theorem: Give a sufficiently large number of monkeys keyboards....gently nudge them in the right direction with a few banana tokens...let them type away with just enough randomness for an infinite amount of time.
Eventually, they will almost surely produce any mathematical proof imaginable...
Why the snarky headline? If parrots can generate statistically plausible proof snippets, and they are backed by a combination of Lean, other proof techniques stolen from ACL2 and a human mathematician they can still be stochastic parrots.
Aaronson is too close to the "rationalist" cult. Maybe he should sit down with ACL2 and get a clue. Emily Bender didn't speak about all that augmentation when coining the phrase.
I don't understand what this objection is even trying to argue. LLMs aren't intelligent, because they're capable of tool use?! One of the few other things we use as an example of how humans are more intelligent than most other animals?
Isn't it obvious? Scott Aaronson is a theoretical computer scientist, which is essentially a mathematician. Scott is likely seeing his career prospects disappear in front of his eyes just like so many others.
Too many people only see the perils of something once it affects them.
The optimists think AI will be like the cotton gin where making something cheaper and more accessible actually increased the demand for slaves as the price dropped. I'm more pessimistic. A lot of jobs are going to collapse in demand in the coming years. For awhile those at or near the top will survive and possibly make a lot of money but the rising floodwaters come for everybody.
> Scott is likely seeing his career prospects disappear in front of his eyes
Why would he be seeing that?
The combination of a mathematician and lean is powerful, and the combination of a mathematician, lean, and AI is more powerful still. Note that the mathematician is always part of that picture. That is fundamental -- no software system, no matter how big and complicated, can actually solve every math problem (this is a fundamental result in math that started the entire field of computer science). If you read TFA he basically tells you what he is actually worrying about here:
"the Lean formal verification system still has issues, from kernel bugs (which have, in the past, allowed proofs of 0=1) to uncertain soundness of the underlying type theory. In my view, these issues can and should urgently be fixed! Until they are, a sufficiently motivated AI could almost certainly exploit them to prove a false statement!"
He is worried about the proof checking system and the possibility that an AI might exploit bugs in the proof checker, not the disintegration of his entire career because AI can "just figure it all out" or whatever.
> A lot of jobs are going to collapse in demand in the coming years. For awhile those at or near the top will survive and possibly make a lot of money but the rising floodwaters come for everybody.
Funny, I'm actually optimistic for this. Let the flood wash away this messed up system that puts people in tiers and says people should only have access to what they can "afford", not to what they need. And then allows a few to be able to afford almost anything, and many to be able to afford nothing, just by being the first to X, or acquainted with Y.
> Scott is likely seeing his career prospects disappear in front of his eyes
...no he's not, that's a ridiculous statement. He is a prestigious professor with tenure at a top university.
Also, mathematics is much much more than just proving theorems. As the word goes, "Good mathematicians prove theorems, great mathematicians make conjunctures, the greatest mathematicians build theories". Someone has to carve out domains of interest to humankind. And humans will always know better what is of interest to humans than machines. These kind of reactions are a form of AI psychosis.
Are you saying we should address llms with more respect now that their good at making lean proofs? Maybe we could call Claude Dr. Claude and Gemeni could be Professor Gemeni.
I'm 99% sure it's not his actual opinion, but a sarcastic remark aimed at the people who claim that LLMs are just stochastic parrots. Like "your peaceful dog just bit me".
I don't think it counters the boosters to ascribe a result like this exclusively to the "stochastic parrot" though. There's an army of people behind each one of these results.
I think this approach and conclusion applies to a lot more than the immediate context of that conversation.
I can say god exists because 5>3. No matter how valid my claim is that five is in-fact greater than three, if I can't convince people of the connection it's a valid proof of something, just not what I say it is.
It might be true that if the Riemann hypothesis is false, leaves cannot fall to the ground. That would mean a falling leaf demonstrates the truth of the hypothesis. Without showing people why it must be the case in a way they understand it falls short of a genuine proof.
Past performance is not indicative of future results. Airliners are still subsonic after a 100 years of aviation.
Yeah, but that's a choice made for, arguably, very good reasons:
- Noise reduction
- Fuel efficiency
- Reduced pollution and lower environment impact
- Reduced cost of materials, equipment, and maintenance
- Passenger affordability
It's not that we can't make supersonic airliners: it's that it doesn't make economic or environmental sense to.
Gulfstream and Sukhoi even had a partnership to develop a supersonic business jet (which, if you look at the mockups, might be one of the coolest looking planes ever conceived) but they gave up after spending a fairly enormous amount of money already because it was going to cost an absolute fortune to bring to market... and there was no real market for it.
It's not we can't make compute much more denser, there's a new standard informally called as megawatt racks. But deploying them at gigantic scales bring in massive problems as well, so we can keep total energy envelope at a fixed range.
...for the sake of the reasons you outlined.
That is a deliberate decision motivated primarily by economics and secondarily by environmental concerns. In fact airliners today are somewhat slower than they were a few decades ago because the lower speeds are more fuel efficient. We certainly could build fleets of supersonic airliners if it made sense to do so, it is not as if we have no idea how to do it and in fact we built supersonic passenger jets once upon a time.
In general, past performance is actually indicative of future results. It's not definitive, and some phenomenon are random, but the trajectory of technology over time is not some random fluke that just has coincidentally trended by random chance.
There was a period in aviation development when it was widely assumed that speed will continue to rise. Just like in the development of CPUs there was such a period, quite recently.
Sure you're going to get a lot of math problems solved, but food distribution is still going to be stuck by warlords and oligarchies.
Think of it like highway designers and the "one more lane" gridlock delusion. In this case, it's more will smith eating speghetti than cheap, low cost cancer cures.
I think robot wives are also coming - the market will be absolutely massive.
I do not see what this kind of post offer here when there is already at least one thread to discuss Alpöge's result with 1000+ comments.
The feeling of a sneer comes only from the absurdity of the belief being definitively revealed.
For anyone who feels sneered at: the problem is the absurd belief, not the reveal.
"Don't sneer at the Emperor for parading around with nothing on!"
His 9-year-old son is more intelligent than you are. I'm not even surprised at this point.
Eventually, they will almost surely produce any mathematical proof imaginable...
Aaronson is too close to the "rationalist" cult. Maybe he should sit down with ACL2 and get a clue. Emily Bender didn't speak about all that augmentation when coining the phrase.
Now complain about upper case, which is the most likely continuation for a human parrot. Or the new account.
Too many people only see the perils of something once it affects them.
The optimists think AI will be like the cotton gin where making something cheaper and more accessible actually increased the demand for slaves as the price dropped. I'm more pessimistic. A lot of jobs are going to collapse in demand in the coming years. For awhile those at or near the top will survive and possibly make a lot of money but the rising floodwaters come for everybody.
Why would he be seeing that?
The combination of a mathematician and lean is powerful, and the combination of a mathematician, lean, and AI is more powerful still. Note that the mathematician is always part of that picture. That is fundamental -- no software system, no matter how big and complicated, can actually solve every math problem (this is a fundamental result in math that started the entire field of computer science). If you read TFA he basically tells you what he is actually worrying about here:
"the Lean formal verification system still has issues, from kernel bugs (which have, in the past, allowed proofs of 0=1) to uncertain soundness of the underlying type theory. In my view, these issues can and should urgently be fixed! Until they are, a sufficiently motivated AI could almost certainly exploit them to prove a false statement!"
He is worried about the proof checking system and the possibility that an AI might exploit bugs in the proof checker, not the disintegration of his entire career because AI can "just figure it all out" or whatever.
Funny, I'm actually optimistic for this. Let the flood wash away this messed up system that puts people in tiers and says people should only have access to what they can "afford", not to what they need. And then allows a few to be able to afford almost anything, and many to be able to afford nothing, just by being the first to X, or acquainted with Y.
...no he's not, that's a ridiculous statement. He is a prestigious professor with tenure at a top university.
Also, mathematics is much much more than just proving theorems. As the word goes, "Good mathematicians prove theorems, great mathematicians make conjunctures, the greatest mathematicians build theories". Someone has to carve out domains of interest to humankind. And humans will always know better what is of interest to humans than machines. These kind of reactions are a form of AI psychosis.
How on earth can the comment you reply to be misunderstood? It even defends Bender.
The boosters are out in force to gloat today.