As a non-mathematics person, this still depresses me so much. It’s like like a kid finding out there is no santa. Or a religious person realizing there is no god. I can’t really quite describe what about this exactly depresses me, but I feel like ground has been taken beneath my feet. Not just this, the whole llm thing. Not because I’m afraid of losing my job or status. Just that the magic is gone makes me feel sad.
Its depressing for me because it means we will lose the knowledge and eventually just be operating completely out of the dark.
Let’s say 3 decades from now there aren’t any people employed in the field of mathematics, what will that mean? Well for starters it means that no one will have the skills to verify any hardcore mathematics outputs from AI. If AI says it solved something, everyone will just have to take it on faith. But it also means that there might not be any more research into mathematics at all in the not too distant future. Who is going to come up with the next problem to solve if there are no human mathematicians?
If it makes a difference, there is also a different, entirely human proof that was as I understand it, complete but not yet thoroughly checked/formatted/paper written, so had not been released yet. After Alpöge announced the AI assisted proof, it was hurriedly announced.
Unfortunately I rarely see people talk about it so it seems it didn’t get very much noticed
I agree. For me, I think it’s because the people who are pushing this by and large don’t give two shits about mathematics for its own sake. They don’t have any reverence for it, they are using this as a hype engine to convince more suckers to pour billions into their failing company.
They don’t care who they hurt along the way, the jobs or careers they destroy, what industries or fields of study they throw into turmoil. It’s the epitome of the, “move fast, break things” Silicon Valley, Capitalist Accelerationalist philosophy.
And on top of that, they got there by farming the totality of human knowledge, research, and open source software without and often against the consent of the authors/creators. Then they used that to create locked-down, proprietary, corporate LLMs and sell it back the the very people who it farmed that output from.
While yes, I was these days remembering how wonderful it was in the Web of my childhood, and then suddenly remembered that what was IRL was just as ugly as social media and all of tech industry now. It’s the Web that was nicer and far more intelligent.
The tech bros are to blame, but what they did was mainly one act - let in the baboons. What should have remained offline, came online. That’s all.
Now on LLMs - that was expected, honestly, after collecting enough resources and power attempting this was the logical next step.
It will bring havoc, but then it’ll get cold. Unless that kind of technologies is only available to a thin portion of the society. I fail to see how with any semblance of market mechanisms that can remain so (remain, because honestly I’m almost certain that before trying to sell that to us, governments and people of Epstein class and such did use them for like a decade).
Honestly, I think CS research has hinted we were going in this direction anyway. Dependently-typed languages like Idris and Agda have as term-inference for a while, it was based on exhaustive type-directed heuristically-prioritized search but it could still auto-complete/fill some non-trivial code, and Roq tactics could also make things that were disastrously wordy read much more simply and they are effectively a form of compile-time function (Elaborator Reflection is probably a better approach, but Roq’s two-language model is much more well-established).
Even without quite so interesting types, there were academic exercises like “Djinn” that could exhaustively enumerate lambda-calculus function bodies given the parametrically polymorphic type signature and a library of standard functions, and that’s been available for 14+ years? In an entirely different direction we’ve had SMT solvers can could be plugged in to a few languages to verify a LOT of properties that can be expressed in first-order logic.
Still, I think there’s plenty of magic, it’s just not in the places we can lay down (enough) axioms around symbolic structure / grammars.
Probably because human thought and intellect are the things that we, as a species, see as differentiating ourselves from any other animals on this planet. And so these intellectual and creative domains feel (and, in the past, were) exclusively the domain of humanity.
Automation is now taking that away.
What do we do when it seems like we, as a species, are on track to render ourselves useless?
I don’t know.
I wonder if this is what old world explorers felt like when the globe was fully mapped and there was no true frontiers anymore. What do you do when the thing that gave you a unique sense of purpose disappears in the face of technology and automation?
What do we as a society do when everyone experiences that at the same time?
A big difference between the explorers in your analogy and what’s happening now is that the exploration of the globe took generations to wind down, whereas AI has gone from chatbot to narrow genius in five years. (I say ‘narrow’ genius because the top experts in most domains still have a better intuitive understanding of their fields, and can still generally choose the path forward better than the machines.)
I keep thinking about the idea of future shock, which was popular in 1970s and ‘80s sci-fi;
a certain psychological state of individuals and entire societies, and a personal perception of “too much change in too short a period of time”.
I guess I never thought human intellect was very special, and when I learned about neural networks, this sort of AI progress felt inevitable. It’s all still pretty mindblowing, but not completely inevitable
230 megabytes of proof. There is Santa out there for you. It’s no good, it’s some piece of slop that gives “ok” on a formal language interpreter. Even if there’s no bug in interpreter affecting that, confirming that those 230 megabytes have any relation to the task allegedly solved is a monumental human work.
As a non-mathematics person, this still depresses me so much. It’s like like a kid finding out there is no santa. Or a religious person realizing there is no god. I can’t really quite describe what about this exactly depresses me, but I feel like ground has been taken beneath my feet. Not just this, the whole llm thing. Not because I’m afraid of losing my job or status. Just that the magic is gone makes me feel sad.
Its depressing for me because it means we will lose the knowledge and eventually just be operating completely out of the dark.
Let’s say 3 decades from now there aren’t any people employed in the field of mathematics, what will that mean? Well for starters it means that no one will have the skills to verify any hardcore mathematics outputs from AI. If AI says it solved something, everyone will just have to take it on faith. But it also means that there might not be any more research into mathematics at all in the not too distant future. Who is going to come up with the next problem to solve if there are no human mathematicians?
If it makes a difference, there is also a different, entirely human proof that was as I understand it, complete but not yet thoroughly checked/formatted/paper written, so had not been released yet. After Alpöge announced the AI assisted proof, it was hurriedly announced.
Unfortunately I rarely see people talk about it so it seems it didn’t get very much noticed
I agree. For me, I think it’s because the people who are pushing this by and large don’t give two shits about mathematics for its own sake. They don’t have any reverence for it, they are using this as a hype engine to convince more suckers to pour billions into their failing company.
They don’t care who they hurt along the way, the jobs or careers they destroy, what industries or fields of study they throw into turmoil. It’s the epitome of the, “move fast, break things” Silicon Valley, Capitalist Accelerationalist philosophy.
And on top of that, they got there by farming the totality of human knowledge, research, and open source software without and often against the consent of the authors/creators. Then they used that to create locked-down, proprietary, corporate LLMs and sell it back the the very people who it farmed that output from.
they aren’t just using it as a hype engine, they are using it because it’s one of the few textual tasks that is easy to confirm and is hard to do.
While yes, I was these days remembering how wonderful it was in the Web of my childhood, and then suddenly remembered that what was IRL was just as ugly as social media and all of tech industry now. It’s the Web that was nicer and far more intelligent.
The tech bros are to blame, but what they did was mainly one act - let in the baboons. What should have remained offline, came online. That’s all.
Now on LLMs - that was expected, honestly, after collecting enough resources and power attempting this was the logical next step.
It will bring havoc, but then it’ll get cold. Unless that kind of technologies is only available to a thin portion of the society. I fail to see how with any semblance of market mechanisms that can remain so (remain, because honestly I’m almost certain that before trying to sell that to us, governments and people of Epstein class and such did use them for like a decade).
Honestly, I think CS research has hinted we were going in this direction anyway. Dependently-typed languages like Idris and Agda have as term-inference for a while, it was based on exhaustive type-directed heuristically-prioritized search but it could still auto-complete/fill some non-trivial code, and Roq tactics could also make things that were disastrously wordy read much more simply and they are effectively a form of compile-time function (Elaborator Reflection is probably a better approach, but Roq’s two-language model is much more well-established).
Even without quite so interesting types, there were academic exercises like “Djinn” that could exhaustively enumerate lambda-calculus function bodies given the parametrically polymorphic type signature and a library of standard functions, and that’s been available for 14+ years? In an entirely different direction we’ve had SMT solvers can could be plugged in to a few languages to verify a LOT of properties that can be expressed in first-order logic.
Still, I think there’s plenty of magic, it’s just not in the places we can lay down (enough) axioms around symbolic structure / grammars.
Probably because human thought and intellect are the things that we, as a species, see as differentiating ourselves from any other animals on this planet. And so these intellectual and creative domains feel (and, in the past, were) exclusively the domain of humanity.
Automation is now taking that away.
What do we do when it seems like we, as a species, are on track to render ourselves useless?
I don’t know.
I wonder if this is what old world explorers felt like when the globe was fully mapped and there was no true frontiers anymore. What do you do when the thing that gave you a unique sense of purpose disappears in the face of technology and automation?
What do we as a society do when everyone experiences that at the same time?
A big difference between the explorers in your analogy and what’s happening now is that the exploration of the globe took generations to wind down, whereas AI has gone from chatbot to narrow genius in five years. (I say ‘narrow’ genius because the top experts in most domains still have a better intuitive understanding of their fields, and can still generally choose the path forward better than the machines.)
I keep thinking about the idea of future shock, which was popular in 1970s and ‘80s sci-fi;
I’m definitely feeling that now.
I, personally, would be totally fine to spend my days playing unproductive games. Just give me UBI.
Honestly, me too. But we both know that ain’t how it’s gonna go down…
I guess I never thought human intellect was very special, and when I learned about neural networks, this sort of AI progress felt inevitable. It’s all still pretty mindblowing, but not completely inevitable
“Damn it… I was hoping to solve that problem one day!”
230 megabytes of proof. There is Santa out there for you. It’s no good, it’s some piece of slop that gives “ok” on a formal language interpreter. Even if there’s no bug in interpreter affecting that, confirming that those 230 megabytes have any relation to the task allegedly solved is a monumental human work.
This is nonsense.
Humans simply have to validate the question. The whole point of Lean and formal proof is to validate everything downstream of that.
The proof is real, and validated.
The monumental human work is reducing it and then understanding it. Although honestly LLMs will do much of this work too.