the AI is on the cusp of AGI except when the topic is all the moral issues around enslaving a sentient being and doubt that an AI god is even possible, then the goal was never AGI at all and you’re silly for thinking that’s still a thing
-
the AI is on the cusp of AGI except when the topic is all the moral issues around enslaving a sentient being and doubt that an AI god is even possible, then the goal was never AGI at all and you’re silly for thinking that’s still a thing
the AI is a transformative tool that creates unique output when the topic is copyright, but if you evaluate the AI’s output and find that it just blatantly copied something from its training set (Servo, SuperTuxKart, gcc, clang, there’s so many examples when you look), then the AI was never capable of transforming anything and you were silly to think it was
-
the AI is on the cusp of AGI except when the topic is all the moral issues around enslaving a sentient being and doubt that an AI god is even possible, then the goal was never AGI at all and you’re silly for thinking that’s still a thing
the AI is a transformative tool that creates unique output when the topic is copyright, but if you evaluate the AI’s output and find that it just blatantly copied something from its training set (Servo, SuperTuxKart, gcc, clang, there’s so many examples when you look), then the AI was never capable of transforming anything and you were silly to think it was
the AI was a grandmaster chess player until it being horrible at chess became a meme
the AI is an incredible mathematician in countless breathless papers but every time you ask it a basic math question it fucks it up badly
the AI is just a tool until you refuse to use it, then it’s a revolution and we can’t afford to be left behind and using it is required
the AI is an incredible boon to open source, as it fractures communities and generates unsalvageably bad code in quantities so massive it can’t realistically be cleaned up
the AI will set you free, but your access to it is gated by a gigantic corporation that’s losing money at every realistic price point they can charge for the thing
the AI is loved by most people until you look at literally any of the polls we’ve done asking people what they think of AI
-
the AI was a grandmaster chess player until it being horrible at chess became a meme
the AI is an incredible mathematician in countless breathless papers but every time you ask it a basic math question it fucks it up badly
the AI is just a tool until you refuse to use it, then it’s a revolution and we can’t afford to be left behind and using it is required
the AI is an incredible boon to open source, as it fractures communities and generates unsalvageably bad code in quantities so massive it can’t realistically be cleaned up
the AI will set you free, but your access to it is gated by a gigantic corporation that’s losing money at every realistic price point they can charge for the thing
the AI is loved by most people until you look at literally any of the polls we’ve done asking people what they think of AI
the AI is a god that can’t spell, and apparently you’re very silly for thinking that might be an issue
-
the AI is a god that can’t spell, and apparently you’re very silly for thinking that might be an issue
the AI is very good at logic, except that if you feed it a mild, original variation of a logic puzzle it fucks it up entirely in a way that instantly reveals it’s only regurgitating answers from its training data. and this is such a problem for the AI corps that they’ll scrape your variation into their training data too and make very certain the AI outputs the right answer, so it looks like they fixed the AI until you try another variation on the same puzzle.
the AI can finally spell strawberry! specifically. with a really long, weirdly angry answer. because they special cased it in the latest model, and it’s literally just outputting a rant from some anthropic asshole.
-
the AI is very good at logic, except that if you feed it a mild, original variation of a logic puzzle it fucks it up entirely in a way that instantly reveals it’s only regurgitating answers from its training data. and this is such a problem for the AI corps that they’ll scrape your variation into their training data too and make very certain the AI outputs the right answer, so it looks like they fixed the AI until you try another variation on the same puzzle.
the AI can finally spell strawberry! specifically. with a really long, weirdly angry answer. because they special cased it in the latest model, and it’s literally just outputting a rant from some anthropic asshole.
the AI is very skilled, but to do anything of value it needs to run some poorly-formed bash on your system that may or may not delete all your files, or create nonsensical files, or do who knows what
the AI is very capable, and that’s why it needs to do a “tool call” to some shit that was designed in the 70s, and it fucks up most of the calls silently in the background
the AI is a security expert, but all the AI corporations keep getting hacked, and all the AIs keep getting hacked with very easy exploits, and all the companies going all-in on AI keep having embarrassing security incidents due to AI
-
the AI is very skilled, but to do anything of value it needs to run some poorly-formed bash on your system that may or may not delete all your files, or create nonsensical files, or do who knows what
the AI is very capable, and that’s why it needs to do a “tool call” to some shit that was designed in the 70s, and it fucks up most of the calls silently in the background
the AI is a security expert, but all the AI corporations keep getting hacked, and all the AIs keep getting hacked with very easy exploits, and all the companies going all-in on AI keep having embarrassing security incidents due to AI
the AI is a boon for code quality, but all of the code is notoriously bad
the AI is a boon for writing code faster, until anyone measures how long things take and finds out it’s much slower unless you’re vibecoding a garbage heap
the AI is a boon for docs, specs, boilerplate, tests, all the parts of the code you don’t care about, until someone who cares looks at any of it, or any of the bits built on those foundations
-
the AI is a boon for code quality, but all of the code is notoriously bad
the AI is a boon for writing code faster, until anyone measures how long things take and finds out it’s much slower unless you’re vibecoding a garbage heap
the AI is a boon for docs, specs, boilerplate, tests, all the parts of the code you don’t care about, until someone who cares looks at any of it, or any of the bits built on those foundations
the AI was very good at art, and in fact had completely replaced the creative process because it replicates what artists do anyway (this is a real thing that real AI kids posted at me), until the general public made it very clear how much we fucking hate AI art
-
-
the AI was very good at art, and in fact had completely replaced the creative process because it replicates what artists do anyway (this is a real thing that real AI kids posted at me), until the general public made it very clear how much we fucking hate AI art
the AI is just a tool, but no successful tool you’ve ever used had a guy with a paycheck insisting you call it a tool, as public sentiment against the thing turned extremely negative. the only supposed tools that seem to act like this are infomercial scams and financial scams
-
the AI was very good at art, and in fact had completely replaced the creative process because it replicates what artists do anyway (this is a real thing that real AI kids posted at me), until the general public made it very clear how much we fucking hate AI art
so anyway, this one time on awful.systems before most of the AI guys got it into their skulls that it’s not a space that’s friendly to them (this took many years), some AI guy tried to prove AI art was good by posting the magic the gathering cards he generated that he claimed were perfect
so the art was bad of course, but more crucially the cards weren’t consistently laid out, and the rules were nonsense text. this guy made unplayable magic cards and tried to pass them off as good. I don’t think someone who does this even plays trading card games.
this was before the AI corps started scrubbing watermarks so I’m pretty certain some of the cards had a corrupted watermark in the lower right portion of the frame
-
so anyway, this one time on awful.systems before most of the AI guys got it into their skulls that it’s not a space that’s friendly to them (this took many years), some AI guy tried to prove AI art was good by posting the magic the gathering cards he generated that he claimed were perfect
so the art was bad of course, but more crucially the cards weren’t consistently laid out, and the rules were nonsense text. this guy made unplayable magic cards and tried to pass them off as good. I don’t think someone who does this even plays trading card games.
this was before the AI corps started scrubbing watermarks so I’m pretty certain some of the cards had a corrupted watermark in the lower right portion of the frame
this one other time on awful.systems an AI guy tried to convince us he was doing things with his code that were impossible without AI, and he gave examples
and all the examples were such basic editing tasks that we started giving vimgolfed command strings and instructions on how to do all of them in a number of basic text editors. this wasn’t even the kind of thing that needs an IDE.
and that’s the thing, isn’t it? for refactors, the LLM is just executing ordinary commands. it can’t do anything you can’t do.
and if you learn the actual tools you’re using (your editor, your toolchain, maybe an IDE if you really need it), you can do anything the LLM does faster and more accurately.
the “AI is just a tool” thing is rich coming from people who never learned to use the tools of their trade at all
-
the AI is just a tool, but no successful tool you’ve ever used had a guy with a paycheck insisting you call it a tool, as public sentiment against the thing turned extremely negative. the only supposed tools that seem to act like this are infomercial scams and financial scams
@zzt And if you ask for a random number between 1 and 30, it will give you 17, then 23, until someone goes and patches for *this specific question*
-
@zzt And if you ask for a random number between 1 and 30, it will give you 17, then 23, until someone goes and patches for *this specific question*
@afilina the surely creative machine that generates the same passwords over and over if you, for some reason, ask it to generate passwords for you
-
this one other time on awful.systems an AI guy tried to convince us he was doing things with his code that were impossible without AI, and he gave examples
and all the examples were such basic editing tasks that we started giving vimgolfed command strings and instructions on how to do all of them in a number of basic text editors. this wasn’t even the kind of thing that needs an IDE.
and that’s the thing, isn’t it? for refactors, the LLM is just executing ordinary commands. it can’t do anything you can’t do.
and if you learn the actual tools you’re using (your editor, your toolchain, maybe an IDE if you really need it), you can do anything the LLM does faster and more accurately.
the “AI is just a tool” thing is rich coming from people who never learned to use the tools of their trade at all
@zzt couldn't agree more with every point in this thread. Thank you.
-
@afilina the surely creative machine that generates the same passwords over and over if you, for some reason, ask it to generate passwords for you
-
the AI is very good at logic, except that if you feed it a mild, original variation of a logic puzzle it fucks it up entirely in a way that instantly reveals it’s only regurgitating answers from its training data. and this is such a problem for the AI corps that they’ll scrape your variation into their training data too and make very certain the AI outputs the right answer, so it looks like they fixed the AI until you try another variation on the same puzzle.
the AI can finally spell strawberry! specifically. with a really long, weirdly angry answer. because they special cased it in the latest model, and it’s literally just outputting a rant from some anthropic asshole.
@zzt Similar to the "strawberry" thing... If you ask Gemini to count to 100, it used to have a hard coded response bragging about how the other models couldn't do that.
-
the AI was a grandmaster chess player until it being horrible at chess became a meme
the AI is an incredible mathematician in countless breathless papers but every time you ask it a basic math question it fucks it up badly
the AI is just a tool until you refuse to use it, then it’s a revolution and we can’t afford to be left behind and using it is required
the AI is an incredible boon to open source, as it fractures communities and generates unsalvageably bad code in quantities so massive it can’t realistically be cleaned up
the AI will set you free, but your access to it is gated by a gigantic corporation that’s losing money at every realistic price point they can charge for the thing
the AI is loved by most people until you look at literally any of the polls we’ve done asking people what they think of AI
-
the AI is on the cusp of AGI except when the topic is all the moral issues around enslaving a sentient being and doubt that an AI god is even possible, then the goal was never AGI at all and you’re silly for thinking that’s still a thing
the AI is a transformative tool that creates unique output when the topic is copyright, but if you evaluate the AI’s output and find that it just blatantly copied something from its training set (Servo, SuperTuxKart, gcc, clang, there’s so many examples when you look), then the AI was never capable of transforming anything and you were silly to think it was
@zzt at this point, I just wish I could get away from it
-
@zzt at this point, I just wish I could get away from it
-
P pelle@veganism.social shared this topic
all the way down, and that’s why wankers love 
it *so much!*