i'm sick of ai being pushed everywhere

[srazkvt]

firefox just had a popup like "heyyy you want me to summarise this page for you ?"

i saw this, gave it the flip, and went in about:config and set to false or empty string about anything under browser.ml

seriously, i enjoy reading things and doing things myself. do not steal those for me. it is nothing but insulting.

[rose]

SO real

ai and llms definitely can have their uses, but none of those are in user facing applications imo
I really can't wait for this hype to die out and everyone to reverse on it, and openai to finally die out because God their whole "chatgpt has logical reasoning!!!!" bullshit is getting annoying
llms are text prediction, there are cases where text prediction is very much useful, a browser is not that place

[srazkvt]

the only useful usecase ive found to LLMs on actual generating text is for text mockups to not be lorem ipsum but instead sound more like reasonable bulshit

it's probably going to be changed anyway, and doesn't need to be truthful or anything, it's still bulshit text

but it is less self similar than lorem ipsum

that's the only usecase i found

and that very much isn't needed in a browser

[rose]

I mean even for that a Markov chain can be used, which afaik is also much more efficient and doesn't guzzle that much power

only useful llm I've seen yet is at work where there are plans to train it on ci logs and use it to anlyze logs
issue is that it's all a vibecoded mess and they're pushing for a noop ci system where problems will be automatically fixed by an llm
which is like.. no??? what manager circlejerk thought that giving it access to the git repos with commit permissions is a good idea

[srazkvt]

I mean even for that a Markov chain can be used, which afaik is also much more efficient and doesn't guzzle that much power

only useful llm I've seen yet is at work where there are plans to train it on ci logs and use it to anlyze logs
issue is that it's all a vibecoded mess and they're pushing for a noop ci system where problems will be automatically fixed by an llm
which is like.. no??? what manager circlejerk thought that giving it access to the git repos with commit permissions is a good idea

fundamentally, an LLM is a markov chain that needs less memory to run on larger scales, and so can run at all on larger scales without having to use a supercomputer

honestly, LLMs are probably more efficient than a markov chains implemented in python. but obviously at the scale the tech world is running them right now, which is on the scale of the entire fucking internet, then of course it will be a disaster when it comes to sustainability

[rose]

oh huh I always thought that Markov chains where better
guess the baseless ai hate got into my mind :p

though whenever I need random text to put somewhere I just open a random Wikipedia page lol it's faster and I get to learn something new !

[srazkvt]

markov chains are viewed as better because they can't run on the same scale LLMs are pushed to.

though iirc the specifics, llms are more expensive to prepare, but lighter to run, while markov chains are the opposite

[lizzy]

imo the worst thing about the AI hype is how corpo friendly it is and how reasonable normies fall for it. back when the hype was shitcoins and NFTs any normal person with a brain understood how that's it's extremely scammy and to stay away from it. but with AI you'll see your friends and relatives fall for the BS and it's just sad to see.

[srazkvt]

imo the worst thing about the AI hype is how corpo friendly it is and how reasonable normies fall for it. back when the hype was shitcoins and NFTs any normal person with a brain understood how that's it's extremely scammy and to stay away from it. but with AI you'll see your friends and relatives fall for the BS and it's just sad to see.

my sister fell into it hard, and she's watching animated ai slop on tiktok

honestly it's just sad