Login
You're viewing the w3c.social public feed.
  • Sep 30, 2026, 1:07 PM

    RE: hachyderm.io/@mnl/116507196473

    Posting this again since despite the mass of confident of “I will never use #ai / #llm / #llms but also I know everything about how they work and everybody using them doesn’t know how to engineer” posts on my timeline triggered me again.

    Try me, I’m willing to spend 4h on whatever.

    If you can confidently state that everybody using llms is brain broken and “lost”, it should be pretty easy to show how broken and bad whatever I produce is going to be?

    💬 1🔄 1⭐ 0

Replies

  • 💬 1🔄 0⭐ 0
  • Sep 30, 2026, 3:27 PM

    @mnl I’m genuinely interested on this, except that the stack constraints put me off as I don’t like any of the allowed languages. This matters because for me success would be a fully generated codebase I can work on without reaching for a model would, as I consider code not only as something that’s executed by a machine, but an entire corpus of knowledge on how to approach a given problem

    💬 1🔄 0⭐ 0
  • Sep 30, 2026, 3:35 PM

    @RosaCtrl i'm happy to use a different stack honestly. I agree on the role of notation / language. I don't read the go very much anymore (I read the API surfaces), but I have a pretty good model of its runtime implications (esp. in terms of concurrency and performance and deployment) that I don't have all that much with python / typescript (or rather, I have been scarred in the past, deeply). A bit like "hey i'm happy to work on arm64, but x86 is a bit more of a struggle, because I'm not sure how the memory bus works" (not the best analogy).

    What stack do you prefer?

    💬 1🔄 0⭐ 0
  • Sep 30, 2026, 3:40 PM

    @mnl in the last years I’ve only written Swift and Lisp-like 😂 But sure we can find a middle ground. Thing is I have notes on a project I’ve wanted to do for a couple of years now, nothing new, I just thing all existing alternatives are wrong. I wonder if I could clean them up, handle them to you, you give it a shot for 4 hours and then I continue on it, without rewriting or heavily refactoring. I would call that a success.

    I used to like JavaScript a lot, I knew a bunch of quirks and the like, but then they added classes and I stopped liking it

    💬 1🔄 0⭐ 0
  • Sep 30, 2026, 3:42 PM

    @RosaCtrl sounds good, last i tried lisp-y stuff (was a big common lisp dev back in the days) it just caused a bit too much hassle with properly giving attention to closing parens (while } ) ] help if along quite a bit).

    I use JS a lot in an interpreter sandbox, and find it quite nice as the malleable "scheme-like" it is. I use react/redux/rtk-query for frontend work, mostly out of habit and because storybook.

    Using an exisitng codebase sounds like a nice challenge since the focus will be on "human communication" and not just technical feat.

    💬 1🔄 0⭐ 0
  • 💬 1🔄 0⭐ 0
  • Sep 30, 2026, 3:54 PM

    @mnl I’ll look them up and handle them. I thought I had them on Apple Notes but I may have moved them to a Markdown file because I don’t find them 😅

    This thing would be a web app, but I’m mostly interested in the storage side of things, for which SQLite would suffice.

    JavaScript would be probably be the best if used as Scheme for the web, but I wonder if you would give Janet a shot? Or would it be to much? I’ve wanted to learn it for a while, and is the language I was thinking of using if I ever decided to start janet-lang.org/

    💬 1🔄 0⭐ 0
  • 💬 1🔄 0⭐ 0
  • Sep 30, 2026, 4:45 PM

    @mnl I’m thinking that targeting JS may be fairer so no goalpost is moved. Is a language that I dislike now, but used to know very well. On the other hand, one of my main problems with models is that the comments they generate are usually awful, as in verbose and say nothing relevant. Maybe you prove me wrong with a well documented Janet codebase?

    Found my “note”. Doesn’t describe what it is though, that’s on my head. Shall I add a paragraph?

    Image attached toot
    💬 1🔄 0⭐ 0
  • Sep 30, 2026, 4:50 PM

    @RosaCtrl oh i really don't mind the janet at all, why not do both. If I struggle too much with janet I'll switch to JS/my usual stack. "Failure" is actually the most interesting to me, that's when I get to work out things and learn.

    Is the project a reading tracker / book catalog?

    I would definitely like more about the "intent", what are you using it for, why, when, how (or do want to use it for).

    💬 1🔄 0⭐ 0
  • Sep 30, 2026, 4:52 PM

    @mnl yes, is an alternative to Goodreads and all its clones that fixes the issues I have with them. So intent should be for human understanding?

    💬 1🔄 0⭐ 0
  • Sep 30, 2026, 5:55 PM

    @RosaCtrl yeah like the technical details and even UX kind of derive from it, and it's much easier to prompt from a position of "intent" rather than go straight into minutae.

    "make me an app to replace goodreads and fix a, b, c" vs "make a react app using rtk-query that has a view with a catalogue sidebar and a panel view using blabla.js"

    which one is going to be the most "effective" at producing something useful? The "engineering" part that is really interesting here, and IMO cannot really be solved by sitting at a computer, LLM or not, is figuring out how to bridge one to the other.

    💬 1🔄 0⭐ 0
  • Sep 30, 2026, 5:58 PM

    @mnl right, but usually I come up with the engineering stuff as I write code, and that would be the gap to be filled here. I do have some technical opinions here though. I’ll come back with them

    💬 1🔄 0⭐ 0
  • Sep 30, 2026, 6:10 PM

    @RosaCtrl yeah this might be a key difference, i've always worked this stuff out away from the computer on paper, and then writing things down was mostly typing. which might be why I might have a different opinion than most? 🤷

    💬 1🔄 0⭐ 0
  • Sep 30, 2026, 6:17 PM

    @mnl yeah, I’ve never seen anyone working like that really. Even when sketching something AFK there’s always stuff that gets revealed by code, so for me it has never been just typing. In the same way that making a list of ideas doesn’t transfer to an entire text in one go.

    Which keeps me interested on this. Will be back with the intent

    💬 1🔄 0⭐ 0
  • Oct 3, 2026, 8:06 PM

    @mnl I'm back, does this make sense? tssm.neocities.org/as-goodread

    Edit: yes there are a bunch of holes. Usually this is when I would start programming. In fact, I couldn't just explain in prose what I wrote as types, which is a nice example of what I mean about programming languages and code as a tool for thought. It comes from the notes in the screenshot, and for me the type definition is so much better than trying to type a clear paragraph

    💬 0🔄 0⭐ 1
  • Oct 3, 2026, 8:39 PM

    @RosaCtrl 100 with you on the types, I do that too. So do you want this to be a personal webapp for keeping track of your reading, giving them scores, allowing translations to be filed as their own works instead of being grouped, and would like to do it in janet?

    ---

    I also thought about this quite a bit because depending on where you are at both in your journey building software / familiarity with LLMs that this might trigger a "what am I even doing / good for" (it happens to me so regularly that I got used to it), which is the _opposite_ of what I want to convey. In workshops I give (used to give), I am pretty good at usually showing to the most skeptical people in the room where the value and what I think ways forward are, but I'm worried this might get lost online / non-collaborative environment.

    I was thinking of a couple approaches to this:

    - happy to share on a meet or a stream and just talk/chat about this stuff and have an easier way conveying what my line of thinking and goals are

    - record what I do and kind of show you how I work these days, using your prompt as a starting point, in 1h increments

    - take your prompt, paste it into an LLM, and walk away (which... sucks, not because I wouldn't do that usually, but because I don't know you and we don't have "trust". If it was a friend and we know how we work and what we are interested in, that would be totally fine, and is indeed what I do with a bunch of friends)

    I know this all sounds a bit cryptic and full-of-it, and I might be, but I'm worried that this will convey the doom, and not the bloom.

    💬 1🔄 0⭐ 0
  • Oct 3, 2026, 9:21 PM

    @mnl well the ambitious idea is to make something social, maybe even on the ActivityPub.

    Regarding what to do with it, I don’t know. I guess whatever you are willing to. I got curious about how much can be done in 4 hours, and the quality of it. I’m not skeptic anymore about models’ usefulness, I do use them daily at work to do mostly analysis of what I’m writing, as well as to get some inspiration or solutions when I tackle anything I lack knowledge. But would I use them to generate code and trust it blindly? No. There’s a prediction that goes on the line of programming becoming a matter of writing natural language and nothing that I’ve seen makes me conceive that as a possibility

    💬 1🔄 0⭐ 0
  • Oct 3, 2026, 10:02 PM

    @RosaCtrl sweet! I have a big ass deadline this week so it’ll probably be until next weekend. Appreciate this very much, and actually something I wouldn’t mind having myself.

    💬 1🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0