Login
You're viewing the discuss.systems public feed.
  • dasdom 🏳️‍🌈dasdom@chaos.social
    May 5, 2026, 7:19 PM

    @helge In the blog post I try to describe that I *need* the boring stuff and the repetition to have the idea in the first place. There is no shortcut.

    💬 1🔄 0⭐ 0

Replies

  • Helge Heßhelge@mastodon.social
    May 5, 2026, 7:32 PM

    @dasdom I usually do it the other way around, I literally spend months (for some things even years) working out or understanding a thing in my head. How it should work, how people would use it. And only if the idea is in good shape, I start coding it. That part is mostly mechanical. And doing that myself is mostly more efficient because transferring my head to someone else (either human or machine) would be unreasonable, transfer speed wise.

    💬 1🔄 0⭐ 0
  • Johannes Brakensieklazarus@fosstodon.org
    May 5, 2026, 7:58 PM

    @helge @dasdom The crucial point maybe is that you learned to think in a way that it can be described in a certain programming language very well. The promise of LLM is you can disconnect the content of your thoughts and their language from the technical language in which it needs to be described to become functional. I find that a haunting idea. Question is if this is going to work in general.

    💬 1🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Johannes Brakensieklazarus@fosstodon.org
    May 5, 2026, 8:01 PM

    @helge @dasdom So maybe programming is going to work well using LLM, but for finding real solutions it doesn‘t matter whether you use LLM or direct technical language, because you‘ll need to describe a problem‘s solution in a technical way a computer can interpret.

    💬 0🔄 0⭐ 1
  • Johannes Brakensieklazarus@fosstodon.org
    May 5, 2026, 8:04 PM

    @helge @dasdom To continue thinking: Maybe what LLM are pointing at is we need to find easier ways to describe technical solutions, because coding takes too long and using LLM is unbearable inefficient regarding current resource usage.

    💬 0🔄 0⭐ 0
  • Helge Heßhelge@mastodon.social
    May 5, 2026, 9:07 PM

    @lazarus @dasdom I don't think coding takes too long. The reason why LLM's work is that they are trained with coding that took long.
    The problem is that most software engineering codes things that have been already coded hundreds if not thousands of times. The state of the art is an extraordinarily waste of human brain resources. OpenSource was one way to attack the issue (people would just reuse stuff others build already). AI seems to be good at this too. We'll see.

    💬 1🔄 0⭐ 0
  • Johannes Brakensieklazarus@fosstodon.org
    May 6, 2026, 5:05 AM

    @helge @dasdom I agree, libraries and frameworks as code reusage are a solution to this problem. For LLM I see too many new problems they introduce to consider them a solution.

    💬 0🔄 0⭐ 0