Login
You're viewing the front-end.social public feed.
  • jonny (good kind)jonny@neuromatch.social
    Apr 3, 2026, 12:35 AM

    @aredridel @mnl right, yeah. the thing i both love and find to be extremely sad is that one of the biggest things that LLMs do, especially for amateurs and hobbyists, is translate natural language into a set of calls to existing tools. it's awesome that people who wouldn't know what to look for can do that, but it's a real shame when it gets presented as "look what i made" rather than "look at what everyone else as made and isn't it amazing that i can just use it for free"

    not saying that's what's happening here, speaking more generally from what i've seen with people who have never programmed before picking up the tool, and instead of being introduced to a community of open source devs who have been working to make things for free for everyone for decades, the person is only exposed to the oracle and that remains the only point of contact or exposure.

    💬 1🔄 2⭐ 1

Replies

  • mnl mnl mnl mnl mnlmnl@hachyderm.io
    Apr 3, 2026, 12:41 AM

    @jonny @aredridel i think that's the uncanny valley we're in now, until people are like "cute stick figure, yo". We're colalborating with a friend and what we now "share" is basically ideas. He showed me his little tool to walk him through wiring / pinout instructions when making circuits by hand, and I was like "oh why didn't i think of that, can I borrow it" and give him credit for the idea, not the code.

    It's hard for our generation to realize that code as such has no big value anymore, when before it was (if a very problematic) a rough measure of effort and thoughtfulness. I don't know where things are going to go from there, and what awaits us... I picture myself as a punchcard programmer being confronted with a personal computer, and thinking "oh now I can punch so many cards and be done at 10am for the nightly batch run", not realizing that the future are things like CICD and worldwide collaboration and microcontrollers.

    💬 2🔄 0⭐ 0
  • mnl mnl mnl mnl mnlmnl@hachyderm.io
    Apr 3, 2026, 12:50 AM

    @jonny @aredridel I also think it's easy for us programmers to miss _what_ people make, because we are so focused on the technical aspects of it. "it's just CRUD" is so far from what a whole swath of software actually represents, and much care and effort it deserves. I always maintained that programmers are absolutely abysmal at vibecoding, because instead of prompting "i want an app to sort my recipes by saltiness/fattiness", they'll prompt "make a postgres backed python app using flask that uses redux and react for the frontend, with an openAPI gate way, store the schema using zod.js a..." and it's like "yo your prompt doesn't even mention the word recipe".

    I think at that point, once people are proud and happy (maybe rightfully, maybe not), to either assist them by taking care of the more complex part, or sharing our knowledge. Certainly the opensource community on mastodon doesn't make it very welcoming for people who just vibed their first app. But the LLM communities I'm part of are usually very welcoming and encouraging.

    💬 1🔄 0⭐ 0
  • jonny (good kind)jonny@neuromatch.social
    Apr 3, 2026, 12:52 AM

    @mnl @aredridel
    i'm extremely happy to help amateurs learn and to help them out with their code. I don't think it's a fair thing to say that "all programmers are mean to amateurs" (though many are and i think that sucks) but you have to acknowledge what you're asking them to do - in many cases vibe coders are asking people to spend time for free fixing their problems that they themselves did not spend any time to understand, and the problem is really that time spent helping someone with a vibe coded app does not actually empower them in the future (i love to teach!), but more serve as fodder for their LLMs.

    💬 0🔄 0⭐ 3
  • jonny (good kind)jonny@neuromatch.social
    Apr 3, 2026, 12:50 AM

    @mnl @aredridel
    well, i have experienced that "sharing at the ideas level" a lot and then when i look inside the code actually doesn't do anything like what the "ideas" say.

    I will refrain from giving my opinions on the aptness of analogies like LLMs being just like such and such prior technological development, but I do not think the problem is that we need to adapt to code being meaningless, that's the marketing line.

    💬 1🔄 0⭐ 1
  • mnl mnl mnl mnl mnlmnl@hachyderm.io
    Apr 3, 2026, 12:58 AM

    @jonny @aredridel this doesn't to me sound like really something related to LLMs though? And when I say code being meaningless, I mean it in the sense that assembly is "meaningless" (I do a bit of assembly here and there, but it's not my main mode of expression anymore when talking about computation).

    What used to be "nebulous knowledge" (programming methodologies, SDLC, even design patterns or other approaches to complexity/decomposition) is now in urgent need of formalization (in the sense that formalizing / creating notation helps us communicate about things as humans, the same way we had to evolve notations like high-level programming languages from cobol to elixir as our expressive capabilities grew ). I have some sense of what that might look like, and the mathematical community already has some interesting concepts ("tactics").

    The speed at which it is happening is definitely disorienting.

    💬 1🔄 0⭐ 1
  • jonny (good kind)jonny@neuromatch.social
    Apr 3, 2026, 1:10 AM

    @mnl @aredridel again i don't necessarily want to engage in "the entire discussion about all of what LLMs mean" just now because i am on the clock, but yeah source language -- compiler -> compiled assembly is a deterministic operation that is not similar in kind to ideas -- LLM -> code.

    💬 2🔄 0⭐ 1
  • 💬 0🔄 0⭐ 1
  • mnl mnl mnl mnl mnlmnl@hachyderm.io
    Apr 3, 2026, 1:19 AM

    @jonny @aredridel i agree. I was referring to the fact that programming languages were always made for communicating amongst humans, and so is talking about "prompting" or whatever this new way of creating ultimately deterministic machine instructions is. and certainly there are ways of building software LLMs that for all intents and purposes are reliable enough to be built upon.

    building software was never deterministic, and for me personally, even the deterministic code I write by hand is highly underterministic, I have absolutely _no idea_ how it gets executed, what all the dependencies and abstractions I build upon do, and instead use "signifiers" and mental shortcuts to make a call of judgment on something being good enough or not. That part hasn't really changed all that much, except I can now more quickly build tools to help me observe and instrument my system or answer certain questions I have.

    💬 0🔄 0⭐ 0