every ai guy says "i carefully reviewed everything!" but then u actually read the code and it's dogshit
every ai guy says "u can't argue the quality; new models actually are genuinely good" and then u actually read the code and it's dogshit
every pro ai discourse relies on the idea that u will never actually go and look at the results yourself and see that they're shit
every ai booster's argument is "this slot machine that kills the world makes me feel good, so if u make me feel bad about it u must be wrong"
it's industrial scale gish gallop
@jacqueline i am so, so tired

@shuppy @jacqueline it's the reason why all this is temporary.

but it really makes you doubt yourself. strangely.

(i just hope we don't take from this that we should never doubt ourselves, like the G.I. generation in the 70's who sent boomers to hell to "teach them a lesson")

@jacqueline Gaslighting on an unprecedented scale.
@jacqueline I want them to feel bad
@jacqueline The parallels between AI use and pet ownership are a bit scary - both exist mainly to make people feel good about themselves
@rancoisse idk he seems satisfied with the arrangement
@rancoisse @jacqueline a key difference being that more people should treat their pets as sentient beings with needs and motivations, and fewer people should treat LLMs as sentient beings with needs and motivations

@jacqueline I've been thinking about that "feeling good" part a lot lately...

I wonder if folks like it because it makes everyone feel like a middle-manager managing an engineering team.
Like you go from grinding out code yourself to spelling out high-level design decisions and constraints for your "team" to work on and then kinda sit back and supervise their work.

Maybe makes you feel more important/powerful/valuable than before. So every attack against the premise of agentic LLM development becomes a personal threat to your imagined new position and status?

@wall_e @jacqueline
Engineering team --> servitude. I think the feeling into GenIA is the feeling of managing servitude

@Tiempo
As I recall, the entire concept of management and most of its principles originated in chattel slavery. So "managing servitude" is a bit redundant, historically speaking.

@wall_e @jacqueline

@Snapai @wall_e @jacqueline

Yeah, sure. And Linux is actually GNU/Linux and so on. But you know, uses and customs 🤷‍♀️

@wall_e @jacqueline it makes people feel like they have a servant. That's all. It's fulfilling the colonial fantasy of slave labor for everyone (who matters)
@SallyStrange @jacqueline considering shit like OpenWhip or BadClaude exist...yeah, you might be spot on
@wall_e @jacqueline idk. to me, the "fantasy" of being middle management is having agency and getting paid well. Not just being responsible for a bunch of doofuses. Using AI "agents" while still being paid the same and just being told what to do like before doesn't really give me that "fantasy"...
@wall_e @jacqueline I haven't thought about "AI" addiction as somebody living out their power fantasies yet but I guess I can imagine that this is how it works for some. What your comment reminded me of was the notion of: if you discover that you have a different hobby than writing code then please stop writing code and do something else with your life that is meaningful. If you can use "AI" then you are usually privileged enough to be able to choose...
@jacqueline I truly, deeply wonder what is wrong with me that it doesn't make me feel good. Ever.
@jacqueline yeah, gambling addicts say the same except they're much safer for the environment.
@jacqueline kinda like stage magic tricks?
@jacqueline "yeah but it works!" (only if you don't touch it)
@jacqueline should have said impressive, they are genuinely impressive. in their way. https://social.treehouse.systems/@dysfun/116669622500374568
gaytabase (@[email protected])

Content warning: slop cods gonna slop

Treehouse Mastodon

@jacqueline All the extra unicode characters, bogus variables that never do anything, and strange extra packages it seems to really enjoy...

make it faster?  
Make bigger numbers so execs think it's cool?! 🏆️

@jacqueline if you go to the trouble of reading the code you can improve it easily especially if it is dog crap, more power to you
@jacqueline The response to the statement “I tried AI last week and it sucks” is always “Ah, but that was last week, the model that came out this week is amazing”
@jacqueline but what a good dogshit it is! Previously it was all bad dogshit.
@jacqueline ahh, but did you say "make no mistakes"? Common gotcha tbh
@jacqueline
"I only had the AI write the tests"
"So, are you even checking anything?"
"The tests pass"
@jacqueline Because by "I" they mean "the LLM," and by "carefully reviewed everything" they mean "processed the code as an input and produced an output that reinforced my assumption for how everything inside the impenetrable and unobservable black box is working."
@StarkRG well u see i opened the diff up in the github pr viewer, and i scrolled through it slowly, and i read most of it. and it looked fine to me. and i consider myself a pretty thorough reviewer. so yeah, i reviewed it thoroughly.

@jacqueline @StarkRG

How dare you read me like this. 😆

@jacqueline I (claude) carefully («make no mistakes») reviewed (generated some report) everything (what could fit in context window)
@jacqueline "I even added tests (gen by ai ofc), so that proves code does what's intended!"
@jacqueline They read the code but did they understand it which is also dogshit.
@jacqueline the worst part, they might have actually done that and did not realize themselves... :(
@jacqueline If you know a few more AI guys you might level-up to "I let another agent review everything". You still might not want to smell that code. (Source: "trust me"). #agentic_coding #slop

@jacqueline they prefix it by "it's my work, AI assisted"

But then can't even answer simple questions.

@jacqueline Doesn't mean they didn't review it. If they were competent, they wouldn't need AI to spew it for them.