542
Most programmers just google it anyway (lemm.ee)
submitted 1 year ago by to c/programmer_humor@programming.dev
50 comments fedilink hide all child comments
 
you are viewing a single comment's thread
view the rest of the comments
[–] 33 points 1 year ago (7 children)

When I used to do image and signal processing for embedded systems in C and C++, AI was useless. Now that I do backend web development in Python and Ruby, AI is better than me. It really depends on the problem area and how many sample code and answers are out there for it to steal from.

  • permalink
  • fedilink
  • source
  • hideshow 14 child comments
    •
  • [–] 31 points 1 year ago (1 child)

    I do backend development in PHP and Ruby, and AI sometimes has a suggestion that helps me out but is often completely, utterly useless, especially at actually coding the thing from scratch.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 15 points 1 year ago* (1 child)

    Yeah lol, a lot of "I need to do X" and I often get "use functionThatDoesX then" and sometimes it's a wonderful discovery, most often though it just doesn't exist lol

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 10 points 1 year ago

    I asked gpt for code to aim a heliostat

    It needed a module to get the sun's position, it used sun::alt:: azimuth which doesn't exist rather than Astro::Coord::ECI::Sun

    It needed a module to calculate mirror angle between the Sun's altitude and azimuth and the target altitude and azimuth. It left that commented out rather than selecting the altitude halfway between Sun and target and azimuth between Sun and azimuth

    It turns out there's precious little on the internet on how to aim a mirror, partly because it's not popular, partly because it's dead simple

  • permalink
  • fedilink
  • source
  • parent
  • [–] 19 points 1 year ago* (1 child)

    Can confirm. AI is worse than useless for embedded systems.

    I was working with Yocto on a very specialized xilinx chip. I had chatGPT makeup a chapter that didn't exist (in a manual that did), reference non-existent paragraphs from said chapter, and then argue with me quite confidently that that chapter was real and the information it was giving me was accurate.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 4 points 1 year ago

    And as soon as you enter corporate stuff, LLMs are useless again, because most things are integrated into existing ecosystems which LLMs don't know and/or libraries are only used for closed source code.

  • permalink
  • fedilink
  • source
  • parent
  • [–] 3 points 1 year ago

    Really? I had an app that would autogenarate time sheets for work in Google Sheets. I decided to minimise API calls by doing a single call to Google Drive then parse the HTML and reupload. Not a big Python project but ChatGPT hit a wall pretty fast on that one. Though, tbf the documentation was suprisingly opaque so I suppose that goes back to your point.

    That project also produced my finest pile of spaghetti code since I had to account for stretched cells in the HTML parsing. I still have a piece of paper with my innumerate math scribbles. The paper makes sense to me. The code does not.

  • permalink
  • fedilink
  • source
  • parent