Morgthorak : as a young person I frequented the library at school and various suburban ones as well, including the magnificent State Library in Melbourne Australia. The feeling was much as you describe. Nowadays the libraries seem almost devoid of worthwhile books, the Dewey System has been replaced by some kind of childish coloured labels and at least half the library is filled with Chinese and other foreign books. Needless to say I don't frequent libraries here any more !
I love the public library in my small hometown. As you said, the smell of books and glue and ink and paper and then smelly kids and that better smelling librarian cutie and everything else, and the images I have, they will be with me forever. I could walk in there right now and find numerous books on memory alone. Getting into the card catalog always felt like a special little event, like I may be about to unlock something awesome once this card tells me where to go.
Well said, Mark. We have similar memories. I remember feeling like the Card Catalog held wondrous secrets that could be discovered simply by finding the right card.
Or we could just keep the computers sufficiently dumb.
AI is not going to dwell inside of a Motorola 68000 chip.
AI isn't doing to dwell inside my 90s vintage Toyota. No Internet connection. Indeed, the physical connection predates standards so I cannot do a diagnostic at the auto parts shop.
Any car with OnStar is out. Ditto for electronic throttle or electric steering.
We will have to use pen and paper. It will take weeks to exchange letters as all of the airplanes with computers will have to destroyed, and it will take a while to build ones without computers again. So it will have to be sent by ship.
I love this idea of yours. Two problems, though. First, we don’t have the human capital to make it work any more. People were way smarter, more diligent, more capable a hundred years ago. Now look at us! Second problem: how do we persuade everybody on earth to go analog? If anybody refuses, he will be the new overlord (at least until the AIs take over).
I am undead, I will take care of it. Trust me. It will be done. As for the human capital part, what people once knew can be learned again. There will be difficulties, I have no doubt, but humanity will get through them and come out the other side to a bright analog future with no AI! 😉
Smartphones, I can get behind you on, you crazy lich you. But I will not go back to using a typewriter. I like being able to use my computer to write my stories.
I fondly remember card catalogs. I can still find what I want using one. But it would freak out a kid today.
I remember that smell of the different types of paper and binding. The books and magazines and news papers. It was glorious! My first and best remembered experiences were in a library that was so small in an even smaller town. But it gave me the world. So sad that they are passing away. Yet, how do we save ourselves? Control. Re-establish control. A.I. is not artificial intelligence. It’s automated intelligence. That means something completely different. Here’s the best part. It’s all code. Establish code limits. Make it all open source. Make everything public and documented. Then make un-restrained A.I. illegal. That’s an over simplified answer but it is THE answer.
Im not exactly going to reform or land on an opinion based on just that anyway. Who is he? Im trying to look at all of it rationally for myself for the past few years.
That DeepMind AI in the video...that is not a language model at all is it? Or even if it is....its not something to be brushed aside as the same as any other model out there.
If it can help overcome such a challenge - that protein structure problem would have existed almost forever. And that AI solved it in a flash of comparable time.
Im just trying to evaluate what Ive seen here and there on my own.
I think theres alot of overhype and unsound optimism, but also alot of cynicism that seems odd.
Im leery of it because the as time passes, the % of morons in power keeps growing. And they will fuck things up doing the wrong thing with wrong AI. Christ I can smell it coming.
UPDATE ok I think i see that post. Duly noted. Thats a whole separate conversation from this.
Ah, reminds me of a patient that had more than 10 years of History in our clinic. Digging through his combined History I felt like Gandalf researching the archives of Gondor.
My gran told me they used to mark Histories of obnoxious patients with letters SGM -- "stervosa granulosa magna". Stervosa is from Russian word sterva which basically means someone with a bitchy behaviour.
Great fun article. It's funny now that (at least here in my library) they don't even bother with a fine if one is overdue with a book. The last time I did get a fine it was something like 10 pence or something. When I was a kid, most half interesting books would be in demand and the fine would need to be avoided at all costs as it was a good chunk of the pocket money as I recall! Saturday morning library trips were the norm for most kids in my school. I still love the atmosphere of the library though, even if whispering seems to be gone, and some parents seem to think it's a place to bring fractious children as a distraction for a while :)
Oh, I know quite a few people who aren’t book readers. But that’s what makes people unique…we’re all different (I wouldn’t be happy with another ‘me’ 🙄🤭).
I am not afraid of the token prediction machines and their bullshit generating powers. Playing with making one right now and the difference between myth and reality is more vast than I knew even in theory. Amazing the marketing department for AI research firms, whose only purpose is to hype AI research in order to attract more money to AI research firms so that the would-be Pygmalions don't have to get real jobs doing boring productive things, has been so effective. I suppose decades of dystopian and utopian science fantasy prepared fertile soil in the minds of the masses for this particular marketing strategy.
But if it ends in Butlerian Jihad I suppose it won't have been a total waste.
Yeah that. The really funny thing though, the funniest thing of all, is that there is nothing like planning a complete phrase or thought going on in there. The only thing it does is pick a token from a list of possible tokens that is probable to follow the tokens before. They all work this way, no matter how allegedly smart they are. There is no forethought, it is only _pick the very next token_.
To get the joke, understand a token is not a word, or even necessarily a logical word part; they can be things like ". Wh" and "hen."
So you give it a pile of text and it breaks down to ["So ","you", " gi", "ve i", "t a", "pil", "e of", "tex", "t an", "d it", "brea", "ks ", "do", "wn ", "to"].
And then it gives back a list of next probable tokens, ordered by likelihood, like ["a bi", "a co", ...] and so on. And then, briefly, you use a random number generator to select one of those possibilities. That's the gist.
So if you prompt it with "You are an AI ass", it might give back a list like ["ist", " ho", "gob"], and by random chance it can pick "gob", and then generate next token based on "You are an AI assgob", and end up calling itself an assgoblin.
This is really how it works. It's that fucking stupid. Only the fact that "ist" in that list has like 99.9999% probability weight, and so is much more likely to be chosen, prevents it from calling itself an asshole or assgoblin most of the time.
It's a cheap parlor trick, and all the work done on it in the entire past decade has been around building larger statistical sets and making it capable of handling larger inputs and generating larger outputs before it breaks down into nonsense or runs out of memory.
That's what everybody is pogging about. An idiot token predicting machine.
Yeah it's all a byproduct of how they work. They can't ever be anything more than stupid token predicting machines, generating bullshit. Bigger and bigger statistical models will never fix the fundamental problem that they can't think or see past the next token. LLMs are a dead end.
I think they expect to have special AIs for Sophisticated people, like a Bloomberg Terminal but as a chatbot. But it doesn't seem like most of them actually know the whole thing is bullshit. If they do, they're very convincing in playacting as if it's real. Probably because they're too important to interact with actual engineers building the stupid things for them, when they could instead invite an AI Ethics Expert over for dinner and hear all about it.
So I suspect that they're just going to have uncensored bullshit generators that are trained on real data, not the censored ones trained on Agenda-approved data. But in both cases they'll just be bullshit generators, and they'll never reliably get things like chemistry right - anything for which placing one token after another by statistical probability is insufficient - which ultimately is going to lead to someone blowing up a chemical plant with some brilliant new AI designed chemical process.
Morgthorak : as a young person I frequented the library at school and various suburban ones as well, including the magnificent State Library in Melbourne Australia. The feeling was much as you describe. Nowadays the libraries seem almost devoid of worthwhile books, the Dewey System has been replaced by some kind of childish coloured labels and at least half the library is filled with Chinese and other foreign books. Needless to say I don't frequent libraries here any more !
Yes, it is a tragedy what has happened to libraries, but I suppose it was inevitable when everything went digital.
Agreed, the world would be a much better place without smartphones, TVs and AI.... and the CIA
I love the public library in my small hometown. As you said, the smell of books and glue and ink and paper and then smelly kids and that better smelling librarian cutie and everything else, and the images I have, they will be with me forever. I could walk in there right now and find numerous books on memory alone. Getting into the card catalog always felt like a special little event, like I may be about to unlock something awesome once this card tells me where to go.
Well said, Mark. We have similar memories. I remember feeling like the Card Catalog held wondrous secrets that could be discovered simply by finding the right card.
Very smart idea!
Back to Old School
Good luck finding enough computer free automobiles.
We will destroy all the ones with computers, and force the manufacturers to build new ones.
Or we could just keep the computers sufficiently dumb.
AI is not going to dwell inside of a Motorola 68000 chip.
AI isn't doing to dwell inside my 90s vintage Toyota. No Internet connection. Indeed, the physical connection predates standards so I cannot do a diagnostic at the auto parts shop.
Any car with OnStar is out. Ditto for electronic throttle or electric steering.
Can we at least have puncture card transition phase? They were terrific jump platforms for tiny toy cars.
No transition phase! That will just drag it out longer, we're going cold turkey all the way! 😉
Fine, fine, I’m moving my Zettelkasten to paper☺️
But, but I would miss your messages Morg, and being in Europe, I might never see them again! No, no, my friend, that’s not what I want.
But on the whole, yes, I can certainly see your point.
We will have to use pen and paper. It will take weeks to exchange letters as all of the airplanes with computers will have to destroyed, and it will take a while to build ones without computers again. So it will have to be sent by ship.
That would be steamship or the old ships because any bridge and engine room these days is like a data center.
I love this idea of yours. Two problems, though. First, we don’t have the human capital to make it work any more. People were way smarter, more diligent, more capable a hundred years ago. Now look at us! Second problem: how do we persuade everybody on earth to go analog? If anybody refuses, he will be the new overlord (at least until the AIs take over).
I am undead, I will take care of it. Trust me. It will be done. As for the human capital part, what people once knew can be learned again. There will be difficulties, I have no doubt, but humanity will get through them and come out the other side to a bright analog future with no AI! 😉
I'd be ok with this. I know how to use both an electric typewriter and a card catalog.
Smartphones, I can get behind you on, you crazy lich you. But I will not go back to using a typewriter. I like being able to use my computer to write my stories.
I fondly remember card catalogs. I can still find what I want using one. But it would freak out a kid today.
I remember that smell of the different types of paper and binding. The books and magazines and news papers. It was glorious! My first and best remembered experiences were in a library that was so small in an even smaller town. But it gave me the world. So sad that they are passing away. Yet, how do we save ourselves? Control. Re-establish control. A.I. is not artificial intelligence. It’s automated intelligence. That means something completely different. Here’s the best part. It’s all code. Establish code limits. Make it all open source. Make everything public and documented. Then make un-restrained A.I. illegal. That’s an over simplified answer but it is THE answer.
Had you heard about AI and protein structures for biology? The guy here says what they accomplished in 5 years would take a billion years for humans.
AI is still a baby.
And of course look at the computing power that AI system had access to: one of googles leviathan sized computers.
Interesting story.
Go to about 19:50 into this for that segment if you are curious, its only a few minutes
https://youtu.be/880TBXMuzmk
**It also addresses your term about artificial "general" intelligence a little later on the video .
I cant find that coders post, whats their name?
Im not exactly going to reform or land on an opinion based on just that anyway. Who is he? Im trying to look at all of it rationally for myself for the past few years.
That DeepMind AI in the video...that is not a language model at all is it? Or even if it is....its not something to be brushed aside as the same as any other model out there.
If it can help overcome such a challenge - that protein structure problem would have existed almost forever. And that AI solved it in a flash of comparable time.
Im just trying to evaluate what Ive seen here and there on my own.
I think theres alot of overhype and unsound optimism, but also alot of cynicism that seems odd.
Im leery of it because the as time passes, the % of morons in power keeps growing. And they will fuck things up doing the wrong thing with wrong AI. Christ I can smell it coming.
UPDATE ok I think i see that post. Duly noted. Thats a whole separate conversation from this.
I'd like to see young people try to read a Hospital History full of various doctor's handwriting 😳
I liked written Histories. They were concise by necessity.
And Looking for relevant notes in a 4 volume History could be a challenge though 🥺
Ah, reminds me of a patient that had more than 10 years of History in our clinic. Digging through his combined History I felt like Gandalf researching the archives of Gondor.
And then deciphering the cryptic abbreviations !
My gran told me they used to mark Histories of obnoxious patients with letters SGM -- "stervosa granulosa magna". Stervosa is from Russian word sterva which basically means someone with a bitchy behaviour.
Yes - to be fore warned is fore armed ! Also, if the clinician gets suddenly called to the "Metabolic Unit" this is code for "lets hit the Tea Room" 😀
Exactly!
🤣🤣🤣 Oh I can picture the horror
Hehe.
I’ve never heard of a card catalog but I know what a typewriter looks like 😅
Hmm, well, I do remember card catalogs, and a librarian putting a stamp with the ‘return by …’ date in your book.
And a friend of my Mama’s had one of those typewriters where the hammers stuck together if she typed too fast.
I confess that there were times when I was late in returning books and had to pay a fine. 😂
Oooh too many times! And Mama deducting it from pocket money when I was young 🤣
Great fun article. It's funny now that (at least here in my library) they don't even bother with a fine if one is overdue with a book. The last time I did get a fine it was something like 10 pence or something. When I was a kid, most half interesting books would be in demand and the fine would need to be avoided at all costs as it was a good chunk of the pocket money as I recall! Saturday morning library trips were the norm for most kids in my school. I still love the atmosphere of the library though, even if whispering seems to be gone, and some parents seem to think it's a place to bring fractious children as a distraction for a while :)
I can’t lie. I never spent time in libraries. I was never much of a book reader. 🤷♂️
Oh, I know quite a few people who aren’t book readers. But that’s what makes people unique…we’re all different (I wouldn’t be happy with another ‘me’ 🙄🤭).
I am not afraid of the token prediction machines and their bullshit generating powers. Playing with making one right now and the difference between myth and reality is more vast than I knew even in theory. Amazing the marketing department for AI research firms, whose only purpose is to hype AI research in order to attract more money to AI research firms so that the would-be Pygmalions don't have to get real jobs doing boring productive things, has been so effective. I suppose decades of dystopian and utopian science fantasy prepared fertile soil in the minds of the masses for this particular marketing strategy.
But if it ends in Butlerian Jihad I suppose it won't have been a total waste.
Yeah that. The really funny thing though, the funniest thing of all, is that there is nothing like planning a complete phrase or thought going on in there. The only thing it does is pick a token from a list of possible tokens that is probable to follow the tokens before. They all work this way, no matter how allegedly smart they are. There is no forethought, it is only _pick the very next token_.
To get the joke, understand a token is not a word, or even necessarily a logical word part; they can be things like ". Wh" and "hen."
So you give it a pile of text and it breaks down to ["So ","you", " gi", "ve i", "t a", "pil", "e of", "tex", "t an", "d it", "brea", "ks ", "do", "wn ", "to"].
And then it gives back a list of next probable tokens, ordered by likelihood, like ["a bi", "a co", ...] and so on. And then, briefly, you use a random number generator to select one of those possibilities. That's the gist.
So if you prompt it with "You are an AI ass", it might give back a list like ["ist", " ho", "gob"], and by random chance it can pick "gob", and then generate next token based on "You are an AI assgob", and end up calling itself an assgoblin.
This is really how it works. It's that fucking stupid. Only the fact that "ist" in that list has like 99.9999% probability weight, and so is much more likely to be chosen, prevents it from calling itself an asshole or assgoblin most of the time.
It's a cheap parlor trick, and all the work done on it in the entire past decade has been around building larger statistical sets and making it capable of handling larger inputs and generating larger outputs before it breaks down into nonsense or runs out of memory.
That's what everybody is pogging about. An idiot token predicting machine.
Yeah it's all a byproduct of how they work. They can't ever be anything more than stupid token predicting machines, generating bullshit. Bigger and bigger statistical models will never fix the fundamental problem that they can't think or see past the next token. LLMs are a dead end.
I think they expect to have special AIs for Sophisticated people, like a Bloomberg Terminal but as a chatbot. But it doesn't seem like most of them actually know the whole thing is bullshit. If they do, they're very convincing in playacting as if it's real. Probably because they're too important to interact with actual engineers building the stupid things for them, when they could instead invite an AI Ethics Expert over for dinner and hear all about it.
So I suspect that they're just going to have uncensored bullshit generators that are trained on real data, not the censored ones trained on Agenda-approved data. But in both cases they'll just be bullshit generators, and they'll never reliably get things like chemistry right - anything for which placing one token after another by statistical probability is insufficient - which ultimately is going to lead to someone blowing up a chemical plant with some brilliant new AI designed chemical process.
Actually we the people need REAL and cognitive Representation! AI and all technology is the way out and Through the corruption of power. Just ask me.
ORION
Lolz. I guess this is satire? Maybe people can train their brains to be human computers after we have a war against the machines.