Linus Torvalds is surprisingly optimistic about vibe coding - except for this one 'horrible' use

Comments

• otacon239@lemmy.world ⁨1⁩ ⁨day⁩ ago

  He hits it on the head. Good for personal projects, terrible for maintained/mainstream ones.

  source

  • Greddan@feddit.org ⁨1⁩ ⁨day⁩ ago

    You can to get drunk/high at home and paint some shit, it might even work out great, but you don’t show up to your job as a painter in the same state.

    source
• nullpotential@lemmy.dbzer0.com ⁨1⁩ ⁨day⁩ ago

  No. Bad Linus.

  source
• DickFiasco@sh.itjust.works ⁨1⁩ ⁨day⁩ ago

  The proof is in the pudding. When/if “vibe code” gets good enough that it can pass a Turing test (be indistinguishable from human-written code), then it will earn its place in the kernel. I think that day is pretty far off though.

  source

  • givesomefucks@lemmy.world ⁨1⁩ ⁨day⁩ ago

    pass a Turing test (be indistinguishable from human-written code)

    That doesn’t pass a Turing test (using the actual mean of a word/definition instead of just making some shit up and then claiming it means what you want it to mean in additional text that makes it so long everyone starts to wonder why you didn’t just type it out in the beginning like it’s just lasted way past the point it should have ended)

    source
• Ephera@lemmy.ml ⁨1⁩ ⁨day⁩ ago

  I am still wondering whether vibe coding will really get folks excited about programming. It is a way to create stuff, so I can see folks getting excited about vibe coding by itself. But making the jump from there to programming seems like it would be frustrating, since you would need to start over with small projects to have any chance at learning the basics.

  And yeah, for now I do not believe that vibe coding will displace programming, because natural language is in many ways just a worse way to formulate logic.
  It’s like going through a translator to talk to someone in another language. It works for simple instructions and to some degree, you can have the translator explain a more general concept, like how to get to a specific place, without you needing provide every detail, but:

  • You can only guess what instructions the translator actually provides.
  • If the result isn’t as expected, e.g. the person doesn’t show up at the described place, then it is nigh impossible to find out which part of the instructions were wrong. Both because you don’t understand the instructions, but because you might not know the way either.
  • And it is just a less efficient way to communicate than when you know the other person’s language.

  source

  • chicken@lemmy.dbzer0.com ⁨1⁩ ⁨day⁩ ago

    But making the jump from there to programming seems like it would be frustrating, since you would need to start over with small projects to have any chance at learning the basics.

    There’s definitely big pitfalls here but I think being able to actually produce software that is useful to you is a better starting point than toy projects that are only for learning purposes. LLMs producing code normally explain what they are doing, and give more detailed explanations when asked that can be supplemented by looking things up separately, which would be one way to learn enough to make your own edits to get around its mistakes and limitations.

    source
• Strit@lemmy.linuxuserspace.show ⁨1⁩ ⁨day⁩ ago

  I’ve “vibe-coded” (AI assisted) parts of one of my projects. I had to port it from SQLalchemy 1.4 to 2.0. My python skills are already fairly low, so that task was a massive undertaking for me.

  So I had AI help me with it. I tested if it all worked and haven’t found an issue yet. The next release of the software will be the real test, it’s where most users gets it.

  My point is, vibe-coding is fine to get you further along if you are stuck on something. But should not be the sole developer when creating and maintaining projects.

  source
• mindbleach@sh.itjust.works ⁨1⁩ ⁨day⁩ ago

  Downvote all clickbait.

  source