35
AI Coding (geohot.github.io)
submitted 1 year ago by to c/programming@programming.dev
10 comments fedilink hide all child comments
you are viewing a single comment's thread
view the rest of the comments
[–] 12 points 1 year ago (3 children)

It's not though. Programming languages, like assembly before them, are deterministic. If you run the same C code again the same environment, it will do the same thing, and altering the code will alter the behaviour correspondingly. It's possible to reason about it. The same does not apply to LLMs. You can't reason about their behavior, when means you can't build anything non-trivial with them. All that is mentioned in the article.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 3 child comments
    •
  • [–] -1 points 1 year ago (2 children)

    Software you create with LLMs is deterministic, because it is the same code as you produce manually. The process of creating it is maybe not, but it is a task of a programmer to review it before publishing it.

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

    it is the same code as you produce manually.

    LLMs do not create the same code that I would, nor do they produce code at the same level that I would. Additionally, LLMs are not deterministic (normally - there are ways to manually seed some but it's rare). Determinism has a very specific meaning. Compilers supporting reproducible builds are deterministic. LLMs producing a different output each time are not.

    it is a task of a programmer to review it before publishing it.

    Tell that to my coworkers. It's honestly insulting the code I have to review and contribute to. Having used these tools myself, I'm better off writing the code myself.

  • permalink
  • fedilink
  • source
  • parent