Conversation

Notices

1. Embed this notice
   tk (tk@f.kawa-kun.com)'s status on Saturday, 28-Sep-2024 01:56:55 JST tk

   Doesn't using an #LLM to create large portions of your codebase mean that you end up with a lot of code that nobody understands unless they put in the time to go through it and learn what it's doing? If so, it's not much different from writing the code from scratch in the long one. :/

   (Yes, I know about boilerplate and do think that's a valid use case as long as you know what the boilerplate is doing.)

   #AI #programming

   • Embed this notice
     Sick Sun (sun@shitposter.world)'s status on Saturday, 28-Sep-2024 01:56:54 JST Sick Sun

     in reply to
     @tk I find the output to be perfectly readable, there's a perception that the code it produces is "alien" but it isn't imo.
   • Embed this notice
     Sick Sun (sun@shitposter.world)'s status on Saturday, 28-Sep-2024 02:01:09 JST Sick Sun

     in reply to
     @tk reading is way faster than writing and the total time to working code is lower.
     
     I'm not exactly trying to defend it, just telling you my experience.
     
     It's sometimes frustrating because it will work really good all day then you give it a particular problem and you struggle with it over and over to get the right output and eventually give up. I'd say that happens about one out of ten times.
     
     The real advantage is sometimes I literally just can't figure out the right or best way to do something and I ask it and with some back and forth get the answer. I had a problem I left alone for over a year and then came back to it and fought with AI for a while but eventually found a solution that worked that I never could have done simply by reading documentation or blogs.
   • Embed this notice
     tk (tk@f.kawa-kun.com)'s status on Saturday, 28-Sep-2024 02:01:10 JST tk

     in reply to

     • Sick Sun
     @sun That's not my point, though. You still have to spend the time to go through and read it to fix any bugs or add any features down the line. If you'd written it in the first place, you'd already know where everything is from writing it.
   • Embed this notice
     Sick Sun (sun@shitposter.world)'s status on Saturday, 28-Sep-2024 02:04:25 JST Sick Sun

     in reply to

     • Sick Sun
     @tk recently I have been having it explain W3C specifications to me and come up with solutions that incorporate multiple ones into one working solution, along with implementing code. It has worked REALLY good for that, probably because W3C docs are very verbose (they are also really long and numerous, hence needing help.)
     
     I've been following up and verifying in docs that it is interpreting them correctly and it does make mistakes but overall it has been a massive time saver.
   • Embed this notice
     tk (tk@f.kawa-kun.com)'s status on Saturday, 28-Sep-2024 02:04:32 JST tk

     in reply to

     • Sick Sun

     @sun Also, if it's trained on code from bad programmers, the code will end up bad as well.

     Kind of like if you search on Stack Overflow and get an answer from 20 years ago that hasn't been the way to do the thing for a long time, then base the code you're writing on that. Too much garbage data out there. :/

     Sick Sun likes this.
   • Embed this notice
     Sick Sun (sun@shitposter.world)'s status on Saturday, 28-Sep-2024 02:07:39 JST Sick Sun

     in reply to
     @tk Yeah so the problem I was having was that I was trying to solve a particular problem using a common programming library. The big problem for AI is that, there are three major versions of that library that are all incompatible with each other, but crypto morons often never explicitly tell you in their shitty Medium article which one they're using they just say "using ethers.js". So the resulting code from the AI is almost always garbage for that library and I just gave up trying to use it.
   • Embed this notice
     Sick Sun (sun@shitposter.world)'s status on Saturday, 28-Sep-2024 02:20:17 JST Sick Sun

     in reply to

     • Steven Sandoval
     @tk @baltakatei you shouldn't be allowed to use it professionally until you proved you can work competently without it.
   • Embed this notice
     tk (tk@f.kawa-kun.com)'s status on Saturday, 28-Sep-2024 02:20:18 JST tk

     in reply to

     • Steven Sandoval
     @baltakatei It also worries me that a lot of people will forget how to write the more difficult code that they're offloading onto code generators (LLM or otherwise). They'll basically become captive customers out of fear of losing their jobs. :/
   • Embed this notice
     Steven Sandoval (baltakatei@twit.social)'s status on Saturday, 28-Sep-2024 02:20:20 JST Steven Sandoval

     in reply to

     @tk Reminds me of the stove analogy in Anathem (2008) by Neal Stephenson (“Part 7: Feral”):

     “We, the theors, who had retreated (or, depending on how you liked your history, been herded) into the maths at the Reconstitution, had the power to change the physical world through praxis. Up to a point, ordinary people liked the changes we made. But the more clever the praxis became, the less people understood it and the more dependent they became on us—and they didn’t like that at all.”