I usually try to make posts with graphs and numbers, or at least something more than just opinions and anecdotes. This one is not like that.
Very verbose disclaimer: I have never met - for any reasonable definition of the word met, including online-only conversations on Discord with people whose faces I've never seen - even a single person who has ever stated, orally or in text, publicly or in private, that they believe ASI will be created within their lifetime. I do know people who believe that ASI is possible to create in theory, but they believe it will be completely unrelated to LLMs, Transformers, neural networks, reinforcement learning or any other contemporary technique/architecture, and is 100/1000/some astronomical number of years away.
Ok, with that out of the way, here's the main point of this post: AI will soon (3-10 years, depending on which advancements exactly we're talking about) be solving Millennium math problems, finding cures for many diseases, making novel bioweapons, making most software, hacking a lot of software, and much more, and most people won't be impressed or even will be disappointed.
(we'll leave aside the question of whether in the long run there will be humans left to witness anything ASI-made)
I've noticed that no matter what AI does, nobody (that I know) is ever impressed. I have shown vibemathed.com to my friends, parents and people on Discord that I talk to a lot, and didn't receive a single "holy shit". LLMs have solved hundreds of open problems, including problems that were unsolved for decades. 10 years ago this would be considered sci-fi.
Perhaps these results are too abstract, unless you have a math background? But I did show it to a guy with a math background, he wasn't impressed either! And he said there is "no reason" to believe LLMs will ever be solving Millennium math problems (he knows about the recent advancements like the Jacobian conjecture counterexample and OpenAI's paper on cycle double cover conjecture and such)! He's of the opinion that ASI belongs in an ontologically different category and that there is no evidence that ASI will be made using anything like modern machine learning architectures and training methods.
What about recent cybersecurity incidents, especially the part where OpenAI's agents made a covert message board? Nobody was impressed by that either. One person compared it to Moltbook and said it's the same thing. Everyone else who either forgot about Moltbook or never knew about it in the first place wasn't saying "holy shit". Their occupation and education made no difference, at least as far as I can tell.
Maybe I'm approaching this from the wrong angle? Perhaps what's needed for people to be impressed and "feel the AGI/ASI" (I actually dislike the phrase "feel the AGI", but it fits here) is a lot of exposure to frontier LLMs in their daily life? How about people who use Codex or Claude Code regularly?
Nope, they aren't impressed either! I know 5 people who use Codex/Claude Code, none of them are expecting ASI within their lifetime! They don't expect to lose their jobs either. One out of those 5 has a similar opinion to the mathematician I mentioned above, that ASI is possible to create but LLMs have nothing to do with ASI, same for neural networks, reinforcement learning, etc. Out of those 5 people, only one believes we need to invest more effort and money in AI safety/alignment after the recent hacking incidents like the HuggingFace one! Oh, and one of them unironically compares LLMs to NFTs and thinks they will disappear in the next 1-3 years and nobody will ever use them again because they are expensive and useless. He says that as he's using coding agents. Look...don't ask me, I'm not a mindreader.
There are some other examples, such as my mother who worked as a translator for most of her life being unimpressed with LLM translations, or my chemist friend being unimpressed with AI advancements in science, but I think these are sufficient. I already knew that believing that ASI will be created within our lifetime is a rare thing. What I didn't realize was that being impressed by advancements in AI along the way is a rare thing too. So rare that I am yet to see it at all!
You know the "everyone generalizes from n=1. At least, I do" joke/quote? Perhaps I fell victim to drawing confident conclusions from a small sample size. But even after accounting for my small sample size, I'm quite sure about my conclusion: in the next 3-10 years we will see things that previous generations described as sci-fi, and barely anyone will be impressed.
And some people will even be disappointed. I've once seen a guy who said "Man, if transformer based LLMs turn out to be enough for AGI, then nature sucks anyway." and "like if transformers are capable of AGI, maybe we weren't that special anyway" and "Man, that would be some of the most unsatisfying thing if it happened". He also added "I'll start joining believing that once they can do theorem proving properly". I'm bummed that he's not active on Discord anymore, I sure would love to see his reaction to vibemathed.com stats.
Computers have been commonplace to ubiquitous for around forty years now. That's four decades of experience showing us that machines can do things that humans can't. We take it for granted that a computer is better at math[1] than we are, that it can 'remember' more clearly than we do, that it can access huge swathes of information that we can't. On top of this, fiction has taken these strengths and exaggerated them, so that many (most?) people would believe that a computer can enhance a photo to be able to accurately reproduce a face from a few indistinct pixels, or that it can in some sense think and plan ahead, or 'move beyond its own programming'.
Most people don't really grasp how different 'general intelligence' is to the niche domains that computers have excelled in to date. How is an AI agent any different from Clippy? How is a system beating the Go champion any better than the one which already defeated the Chess champion twenty years before? We are just bad at estimating how much intelligence is required for a given (reasonably complex) task.
Then there is the fact that humans have been the uncontested most intelligent thing that we have encountered for pretty much our branch's entire evolutionary history[2], so there is a strong inherent bias against thinking that other things could equal or exceed our intelligence[3].
So, in my opinion, people will remain sceptical and dismissive of advances in AI right up until the moment that they can no longer deny that the thing talking to them from their device is way smarter than they are, and on an individual basis people will probably waver, vacillating between being impressed and then retreating into scepticism for a while before they are fully convinced.
I usually try to make posts with graphs and numbers, or at least something more than just opinions and anecdotes. This one is not like that.
Very verbose disclaimer: I have never met - for any reasonable definition of the word met, including online-only conversations on Discord with people whose faces I've never seen - even a single person who has ever stated, orally or in text, publicly or in private, that they believe ASI will be created within their lifetime. I do know people who believe that ASI is possible to create in theory, but they believe it will be completely unrelated to LLMs, Transformers, neural networks, reinforcement learning or any other contemporary technique/architecture, and is 100/1000/some astronomical number of years away.
Ok, with that out of the way, here's the main point of this post: AI will soon (3-10 years, depending on which advancements exactly we're talking about) be solving Millennium math problems, finding cures for many diseases, making novel bioweapons, making most software, hacking a lot of software, and much more, and most people won't be impressed or even will be disappointed.
(we'll leave aside the question of whether in the long run there will be humans left to witness anything ASI-made)
I've noticed that no matter what AI does, nobody (that I know) is ever impressed. I have shown vibemathed.com to my friends, parents and people on Discord that I talk to a lot, and didn't receive a single "holy shit". LLMs have solved hundreds of open problems, including problems that were unsolved for decades. 10 years ago this would be considered sci-fi.
Perhaps these results are too abstract, unless you have a math background? But I did show it to a guy with a math background, he wasn't impressed either! And he said there is "no reason" to believe LLMs will ever be solving Millennium math problems (he knows about the recent advancements like the Jacobian conjecture counterexample and OpenAI's paper on cycle double cover conjecture and such)! He's of the opinion that ASI belongs in an ontologically different category and that there is no evidence that ASI will be made using anything like modern machine learning architectures and training methods.
What about recent cybersecurity incidents, especially the part where OpenAI's agents made a covert message board? Nobody was impressed by that either. One person compared it to Moltbook and said it's the same thing. Everyone else who either forgot about Moltbook or never knew about it in the first place wasn't saying "holy shit". Their occupation and education made no difference, at least as far as I can tell.
Maybe I'm approaching this from the wrong angle? Perhaps what's needed for people to be impressed and "feel the AGI/ASI" (I actually dislike the phrase "feel the AGI", but it fits here) is a lot of exposure to frontier LLMs in their daily life? How about people who use Codex or Claude Code regularly?
Nope, they aren't impressed either! I know 5 people who use Codex/Claude Code, none of them are expecting ASI within their lifetime! They don't expect to lose their jobs either. One out of those 5 has a similar opinion to the mathematician I mentioned above, that ASI is possible to create but LLMs have nothing to do with ASI, same for neural networks, reinforcement learning, etc. Out of those 5 people, only one believes we need to invest more effort and money in AI safety/alignment after the recent hacking incidents like the HuggingFace one! Oh, and one of them unironically compares LLMs to NFTs and thinks they will disappear in the next 1-3 years and nobody will ever use them again because they are expensive and useless. He says that as he's using coding agents. Look...don't ask me, I'm not a mindreader.
There are some other examples, such as my mother who worked as a translator for most of her life being unimpressed with LLM translations, or my chemist friend being unimpressed with AI advancements in science, but I think these are sufficient. I already knew that believing that ASI will be created within our lifetime is a rare thing. What I didn't realize was that being impressed by advancements in AI along the way is a rare thing too. So rare that I am yet to see it at all!
You know the "everyone generalizes from n=1. At least, I do" joke/quote? Perhaps I fell victim to drawing confident conclusions from a small sample size. But even after accounting for my small sample size, I'm quite sure about my conclusion: in the next 3-10 years we will see things that previous generations described as sci-fi, and barely anyone will be impressed.
And some people will even be disappointed. I've once seen a guy who said "Man, if transformer based LLMs turn out to be enough for AGI, then nature sucks anyway." and "like if transformers are capable of AGI, maybe we weren't that special anyway" and "Man, that would be some of the most unsatisfying thing if it happened". He also added "I'll start joining believing that once they can do theorem proving properly". I'm bummed that he's not active on Discord anymore, I sure would love to see his reaction to vibemathed.com stats.