Genuinely wild how many people keep saying "AI is useful you just need to check the results and walk through everything it does" What kind of rinky dink ass work do these people do where verifying every single claim made by an LLM on a project is faster than just doing it correctly yourself lol - ThreadSky

mugrimm.bsky.social • 60 days ago

Genuinely wild how many people keep saying "AI is useful you just need to check the results and walk through everything it does"

What kind of rinky dink ass work do these people do where verifying every single claim made by an LLM on a project is faster than just doing it correctly yourself lol

Reposted from Regime Accountant, CPA

The people defending their use of chatGPT for technical work, I’m sorry but I would fire you.

Comments

okemsrazor.bsky.social•60 days ago

I might as well ask a fortune teller everything. same accuracy

charlotteis.online•60 days ago

lol using the summarize feature only to have to double check the full document anyway in case the ai hallucinated a bunch of ideas

mcbiohazard.bsky.social•60 days ago

If you want me to manage an employee, give me supervisor pay in my scope or GTFO. I don't have time or attention to babysit a coworker if I'm not getting compensated for it.

stablestrategys.bsky.social•60 days ago

fr like if i told you a coworker was “useful” provided you walked them through the entire process of every single task and also had to triple-check everything they said because they are known to enthusiastically lie, you’d say they’re “bad at their job”. why does the computer get a pass?

lydy.bsky.social•59 days ago

I get the feeling that the primary use of AI for corporations is they can deny responsibility for their actions. If an employee tells a customer "this will absolutely not harm you in any way", they are liable. There's a lot of work to make sure that's not true if it's a computer making the error.

extremelyserious.bsky.social•60 days ago

welp, there it is. this is the thought ive been having about this whole thing but have been unable to elucidate myself. good work 10/10

mugrimm.bsky.social•60 days ago

The only real exception we've found is ML training on very rigid domains with A/B testing and verification (think sorting piles) but even that isn't ChatGPT it's a specialized system we built from the ground up with tons of fail safes and no consequences for failure.

iamtherog.com•60 days ago

No one ever talks about the amount of human labor in the Global South is being expended in the name of completely fake AI projects. People always think I’m lying when I say the “self driving robots” are just being steered by some guy in the Philippines with a Game Cube controller. They are!

twohitgood.bsky.social•60 days ago

Mechanical Turks are making a huge comeback.

iamtherog.com•60 days ago

People are mystified when I tell them about this and then I’m like, the literal name of Bezos’ product is MTurk.

mugrimm.bsky.social•60 days ago

But like yeah, if you got any real amount of complexity or content it's basically useless.

progressiveknife.bsky.social•60 days ago

> exciting new ML model
> ask if it's generative or a classifier
> pull out a flowchart explaining what's generative and what's a classifier
> "it's a good model sir"
> it's generative

kevinrohr.bsky.social•60 days ago

i can use windsurf to build an entire app in a few days that would’ve taken me months without writing a line of code. what are you talking about

lenorarose.bsky.social•59 days ago

Every person I've seen kinda praise it for coding is someone who either doesn't know or barely knows the code language. Everyone conversant seems to say it does the same thing it does with English and includes whole streams of nonsense. Or code that only works in a similar but differing language.

kevinrohr.bsky.social•59 days ago

been writing code for 25 years, am highly paid to do so as a consultant.

you can complain about aspects of the code these tools write, or that they occasionally totally fuck up your app or misinterpret your instructions. but to say what they produce is inefficient or useless is totally wrong

kevinrohr.bsky.social•59 days ago

and these tools are in their infancy. they’re good enough now that i’m not sure if anyone is going to be writing much code in 5 years, let alone 10

lenorarose.bsky.social•59 days ago

The only way to "improve" the output is to steal more code from elsewhere and feed it into the machine. And if it scrapes the internet now, it recursively feeds on AI output, so the bad code quotient increases; most improvements actually come from humans manually tweaking it in sweatshops.

kevinrohr.bsky.social•58 days ago

wow you really do not know what you are talking about. you think ai tools are just googling things and copy/pasting code chunks it finds into your code? and you think sweatshop workers are actually writing code?

that’s stunning and i don’t think you should be talking about this topic

kevinrohr.bsky.social•59 days ago

i think a lot of people of my political persuasion (left) saw crypto scams and nfts and also love to highlight flaws with ai tools, implying they’re a scam too.

i don’t think they’re a scam, i think they’re a threat.

lenorarose.bsky.social•59 days ago

I think they're both.

kevinrohr.bsky.social•60 days ago

i’m not even saying that’s a good thing. i don’t actually learn whatever language i’m using, and in 5 years my job might be gone. but to act like ai doesn’t massively speed up development is just delusional

kevinrohr.bsky.social•60 days ago

the biggest adv at the moment is setting up a new project imo. so much time goes into getting all the stupid config right, and this ai thing can just generate the whole project structure and all the config files for me in like 30 seconds.

progressiveknife.bsky.social•60 days ago

Very expensive regex writing machine!

mezentine.bsky.social•60 days ago

People keep talking about its insufficiency in replacing the humanities but honestly I remain deeply skeptical that it would have been useful for much more than a teaching aid for my mechanical engineering degree. Its math performance continues to underwhelm.

mezentine.bsky.social•60 days ago

As a piece of mostly functional (depending on the domain) hypertext with a conversational interface, sure, asking “What does this fluid density term represent” is faster than tracking it down in the textbook and maybe the conversational interface is less intimidating, but it’s still bad at math!

leoastrum.bsky.social•60 days ago

I just went through the replies to that post to block the worst offenders, and a lot of them had very similar numbers of followers/following (usually in the 20-30 range) and (at a glance) a handful of very generic posts. Given how similar they are, they're probably not all real people

cyberpilot.bsky.social•60 days ago

You forget middle management.

verm.me•60 days ago

I think part of the puzzle here is that in domains where it is possible to verify the result (software, certain scientific problems, etc.), that’s already what you should be doing. And if you don’t care enough to verify, then maybe “usually basically right” is good enough.

sapphichazard.bsky.social•60 days ago

Not caring enough to verify is definitively not "good enough" when it comes to software, for what it's worth. That concept is in defiance of the entire concept of a software development lifecycle.

cyberpilot.bsky.social•60 days ago

Twitter feeding users slop would not be fundamentally worse if the slop is pushed by ai. Or generated by ai.

sapphichazard.bsky.social•60 days ago

If? Isn't it already?

cyberpilot.bsky.social•60 days ago

Unsure if anybody there could pull that off, tbh.

sapphichazard.bsky.social•60 days ago

AI is producing a large amount of content for twitter, and is pushing that content, as far as I'm aware. Inasmuch as chatbots and algorithmic elevation count as AI, anyway.

At least I'm under that impression? I haven't been there in a long time. Better things to do than spend time on a nazi site.

bug-deal.bsky.social•60 days ago

there's no end of bots providing no end of generated content, though.

sammisolsdottir.bsky.social•60 days ago

So many folks weren’t immunized with Eliza and it shows

cpi.bsky.social•60 days ago

The secret sauce is that they don't actually verify most/any of the claims and just hope it never comes back to bite them...

Comments

Posting Rules

Reply