807
"Why would I spend time bonding with my kid when I can just get an AI to do it for me?" (europe.pub)
submitted 1 day ago* by to c/fuck_ai@lemmy.world
184 comments fedilink hide all child comments
 
you are viewing a single comment's thread
view the rest of the comments
[–] 11 points 1 day ago (6 children)

I've seen AI rewrite entire code bases overnight that would have taken a team of humans over a year.

And was the code better, cleaner and more readable? Moreover did it actually still do what it was supposed to and not introduce further unknowns?

I get that it's basically just a prediction engine but thinking of it like that is unhelpful in my opinion.

Unhelpful in the sense that it is what it actually is or unhelpful in the sense that it makes you question the fake narrative we're getting shoveled to us by tech bros and traditional mass media?

Like that's just what it is. All I want is folks to be honest about the limitations of this tech and what is underpinning it. It's still very impressive but at the end of the day there will always be an inherent problem with LLM in that they are non deterministic - you can give the same AI the exact same prompt 10 times and you'll get 10 different replies.

There is no way to get around that. You can't have a perfectly predictable output and even 1% of the time error is not good enough for widespread adoption when a mistake could mean someone dies or is horribly injured or loses a ton of money etc etc

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 6 child comments
    •
  • [–] 3 points 5 hours ago (3 children)

    And was the code better, cleaner and more readable?

    I'm going to interject here. I've got family who does software development professionally, and has leaned heavily into AI to assist with coding, and the answer is 'yes', but there are some caveats. If you just tell an AI what to do and let it run hog-wild, you end up with shit. When you have an experienced programmer review and have the AI make targeted changes, it is extremely effective. Some of his projects are seeing 1000% improvements on resource usage and/or compute time.

    Here is the rub though. You're not going to end up with experienced programmer to manage programming AI agents if they never get any actual experience programming because they've just been having an AI do all the coding for them.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 3 child comments
    •
  • [–] 1 point 4 hours ago (2 children)

    Right I appreciate your insight. I'm not heavy in coding myself, I brush up against code and use it for some automation and assistance but I've also seen it write literal nonsense for code.

    That rub is so real. I asked the thing to write some code for me for a Sailpoint workflow and I wasn't familiar with the syntax so figured it'd be a good place to use AI. That thing hallucinated extremely convincing code except it just wholesale made up how it worked and of course it looks fine to my eyes but then go to run it and error -> work to fix that error and then -> error, etc etc.

    After chasing my tail for an hour or two I finally stopped and went to pull up some official documentation and that's when I figured out the AI was just hallucinating like a motherfucker. So I went back and did it the "old fashioned" way, but at least the AI did give me an idea of what I had to do.

    Anyways, I could see someone who isn't me just running around in circles for days because they believe the LLMs we call AI are giving them actual answers.

    I blame some of it on how AI got mixed into web search. Traditionally, the resource from the web would at least be written by a human and usually relevant. Generally semi trustworthy and not made up. Now people see search results with AI summary/answer at the top and that is not vetted, could be a straight up hallucination*, or even just a total misinterpretation of your search phrase so totally irrelevant.

    *I wish this term was different because it has an implication that the AI can and cannot hallucinate when the truth is the AI is always hallucinating. Put another way, to the LLM there is no such concept as truth. It annoys me that people seem to believe that the AI "knows" stuff. It does not. Stop anthropomorphizing the fancy robot.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] -1 points 12 hours ago (1 child)

    It's simply better than a human could have produced in the same time. If it works who cares how clean it is like this is what I don't get about programmers they're so obsessed with cleanness and it doesn't matter it's so unimportant to the end goal yeah I get that maintenance of the program would be a pain but the AI is going to do that so who cares.

    I do sort of feel that this community is so obsessed with hating AI that they don't give it the benefit of the doubt when it actually achieve things.

    The problem with AI is not the product itself it's the way that people use it like let's not lose sight to that, that's not give all the tech Bros ammunition by trying to claim that everything AI does is terrible and bad and awful simply because it was the AI that did it.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] 1 point 6 hours ago

    who cares how clean it is like this is what I don't get about programmers they're so obsessed with cleanness and it doesn't matter

    It matters to the compiler so the code actually will run and not break. It matters to whoever has to fix or update it when things change. It matters to the hardware it runs on and how much memory that hardware has.

    It very much matters.

  • permalink
  • fedilink
  • source
  • parent