Post by @zzt@mas.to

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

8 Replies

18 Boosts

[object Object]

@zzt@mas.to 6 hr. ago

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

[object Object]

@zzt@mas.to 6 hr. ago

the AI is a god that can’t spell, and apparently you’re very silly for thinking that might be an issue

[object Object]

@zzt@mas.to 6 hr. ago

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.

[object Object]

@zzt@mas.to 5 hr. ago

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

~8 more replies

[object Object]

@zzt@mas.to 5 hr. ago

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

[object Object]

@zzt@mas.to 5 hr. ago

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

[object Object]

@zzt@mas.to 4 hr. ago

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

[object Object]

@zzt@mas.to 4 hr. ago

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