Conversation

Notices

1. Embed this notice
   7666 (7666@comp.lain.la)'s status on Friday, 09-Aug-2024 08:45:39 JST 7666
   one of the worlds most punchable faces
   • Embed this notice
     Sick Sun (sun@shitposter.world)'s status on Friday, 09-Aug-2024 08:45:36 JST Sick Sun

     in reply to

     • of nothing
     @7666 @apropos the current craze is you have the LLM write tests and then you have the LLM write the code then you run the tests on the code, and do this in a loop until all the tests pas.
   • Embed this notice
     7666 (7666@comp.lain.la)'s status on Friday, 09-Aug-2024 08:45:38 JST 7666

     in reply to

     • of nothing
     @apropos "what's that? the AI wrote a bug? don't worry about that. Our AI based QA platform will catch it."
   • Embed this notice
     of nothing (apropos@fsebugoutzone.org)'s status on Friday, 09-Aug-2024 08:45:39 JST of nothing

     in reply to
     @7666
     >AI can increase productivity by 20% to 40%
     congratulations, your work is easier now
     >Dell's targets for teams' output aligned with those levels.
     congratulations, your work is the same as before now
   • Embed this notice
     7666 (7666@comp.lain.la)'s status on Friday, 09-Aug-2024 08:47:59 JST 7666

     in reply to

     • Sick Sun
     • of nothing
     @sun @apropos can't wait for it to hallucinate variables that don't exist or that the tests pass but it immediately segfaults because it didn't actually have an execution environment it just went "beep boop running code now" and it hallucinates what a return might potentially be
     Sick Sun likes this.
   • Embed this notice
     Sick Sun (sun@shitposter.world)'s status on Friday, 09-Aug-2024 08:50:42 JST Sick Sun

     in reply to

     • of nothing
     @7666 @apropos I actually moderately use AI to write code and it makes a lot of mistakes you have to manually fix but overall saves time. The hardest part is training yourself not to feel disgust at how ridiculous the process is.
   • Embed this notice
     Sick Sun (sun@shitposter.world)'s status on Friday, 09-Aug-2024 08:52:44 JST Sick Sun

     in reply to

     • of nothing
     @apropos @7666 the logic is that it's easier to write tests than code so the LLM gets it right or closer to right far more often, then you can just automate the code generation process so even if it takes longer, it took less of your productive time.
   • Embed this notice
     of nothing (apropos@fsebugoutzone.org)'s status on Friday, 09-Aug-2024 08:52:45 JST of nothing

     in reply to

     • Sick Sun

     @7666 @sun the appeal of testing is that it’s double-entry bookkeeping for code. You have the desired behavior expressed multiple ways that can be mechanically confirmed to agree with each other. That’s all: these different ways of saying what the program should do, they agree, so the code might just do what it’s supposed to do.

     Faking that with LLM-generated code, which does not in any real sense know what the program should do, is just going to result in obvious bugs that pass tests.

     Sick Sun likes this.
   • Embed this notice
     Sick Sun (sun@shitposter.world)'s status on Friday, 09-Aug-2024 08:56:38 JST Sick Sun

     in reply to

     • Sick Sun
     • of nothing
     @apropos @7666 I want to make clear though that I am NOT a retard and I know good and well that just because a code passes a test doesn't mean it's not the worst code imaginable.
   • Embed this notice
     iced depresso (icedquinn@blob.cat)'s status on Friday, 09-Aug-2024 08:57:30 JST iced depresso

     in reply to
     @7666 > value-added layers
     spoken like a man who has never slung a line of code
     Sick Sun likes this.
   • Embed this notice
     Sick Sun (sun@shitposter.world)'s status on Friday, 09-Aug-2024 08:59:21 JST Sick Sun

     in reply to

     • iced depresso
     @icedquinn @7666 the external research: Gartner shovelware content.
   • Embed this notice
     Sick Sun (sun@shitposter.world)'s status on Friday, 09-Aug-2024 09:07:44 JST Sick Sun

     in reply to
     @7666 I bet you could replace this asshole with an LLM trained to generate corporatese
   • Embed this notice
     7666 (7666@comp.lain.la)'s status on Friday, 09-Aug-2024 09:07:45 JST 7666

     in reply to
     bonus meme
   • Embed this notice
     ベッサーヴィッサー :verify: :verified: :verified_flashing: (besserwisser@shitposter.world)'s status on Saturday, 10-Aug-2024 06:23:35 JST ベッサーヴィッサー :verify: :verified: :verified_flashing:

     in reply to

     • iced depresso
     @icedquinn @7666 I had a job interview today and the CEO was really excited about AI rewriting COBOL code in a different language.
     iced depresso likes this.