@TomSwirly@toot.community
@TomSwirly@toot.community avatar

TomSwirly

@TomSwirly@toot.community

.uk ➡️ .at ➡️.ca ➡️ .nyc ➡️ .nl

Check out my silly little bot here: https://botsin.space/@fortune

This profile is from a federated server and may be incomplete. Browse more on the original instance.

Cloudguy, to random

deleted_by_author

  • Loading...
  • TomSwirly,
    @TomSwirly@toot.community avatar

    @Cloudguy Why would one have an Amazon account in the first place? 😃

    They're an evil company: I haven't done business with them in almost a decade!

    Thanks for the warning for the community, though.

    wogan, to random

    Played around a bit yesterday with the idea of using OpenAI to write software. Turns out it's surprisingly easy to get the model to produce code that will work, it's just the indeterministic nature of the gpt-3.5-turbo model (and/or my naivety in using it) that makes for a challenge.

    It's not much of a challenge, though - after writing 3 short specifications (two related Models and a Service, using Laravel) it produced code that was 80% usable right off the bat.

    TomSwirly,
    @TomSwirly@toot.community avatar

    @wogan I write computer programs for a living and anything to speed it up, but I'm not seeing the use of 80% usable code.

    TomSwirly,
    @TomSwirly@toot.community avatar

    @wogan Code that's 80% right is not 80% of the effort involved. Debugging is twice as hard as programming, says Kernighan.

    I've been writing code for almost 50 years now. I rely on getting the code almost 100% right on the first try.

    All programmers spend very little of their time writing new code and most of it making incremental changes to an existing system that preserves a large number of conditions, some implicit. Getting that 80% right isn't even close to good enough.

    TomSwirly,
    @TomSwirly@toot.community avatar

    @wogan It might be that that day will come. It's certainly not here yet, or even close.

    I don't believe that large language models can do it. I believe at some point of complexity, the coder needs a "mental" model of the system internalized. "What other people wrote in a similar situation," really is only useful for tiny toy programs.

    I've done quite a bit of teaching of beginner programmers, and quite a lot of them approach it like these AI systems do:

    --- more ---

    TomSwirly,
    @TomSwirly@toot.community avatar

    @wogan Which is to say, they don't actually have a model of how their programs work, but they rely on looking at a lot of other code, and completing their project based on that.

    They never get past toy programs.

    TomSwirly,
    @TomSwirly@toot.community avatar

    @wogan

    > Out of curiosity, are you also the sort that never reads industry blog posts, how-to's, Q&A answers etc and always just go directly to the source code of whatever you're working on to understand it?

    I read a huge amount of technical literature, and I read a huge amount of code. It is almost certain that if I use a package I'll read the source, if only for entertainment. The source is the ultimate truth, but life's too short to read all of it always. 😄

    TomSwirly,
    @TomSwirly@toot.community avatar

    @wogan 10,000x more non-working code has no real value.

    I strongly disagree that correct, high-quality code has become a commodity yet. It might happen at some time in the future.

    From my many hours spent with the semantic parrots, there will have to be a breakthrough before we get code that's even as good as a talented junior engineer.

    https://github.com/rec is my open-source code, and also, my best idea of what quality code looks like.

    TomSwirly,
    @TomSwirly@toot.community avatar

    @wogan I don't believe you can just throw a lot of material into a program, and actually get correctness out of it, without some new, as yet uninvented, system that can tell truth from falsehood.

    An LLM simply tries to imitate what other people have done before in the past in a similar situation. The idea that truth and correctness will magically emerge from that, or that these things are unimportant, are not founded on any rational basis.

    TomSwirly,
    @TomSwirly@toot.community avatar

    @wogan Very cool, very interesting!

    I actually had a friend also use an LLM to try to do the same code. It started OK, but it did not lead me to a light turning on. As we went on, it made up symbols.

    Unfortunately, there doesn't seem to be a OpenDMXController library. fade_to_full_brightness doesn't seem to exist as a symbol, at least on github, and there seems little use for such a function. (A fade would be very useful, but "full brightness" is not an interesting place. 😄

    -- more --

    TomSwirly,
    @TomSwirly@toot.community avatar

    @wogan Oh, believe you me, the lighting world would be better off with such a library!

    I ended up settling on a tiny library that just does DMX for this one FTDI chip: https://github.com/generalelectrix/pyenttec/blob/main/pyenttec/

    I found several other frameworks, but they were not good in general, and didn't seem to drive the FTDI chip correctly. That chip is in all the cheap DMX interfaces.

    TomSwirly,
    @TomSwirly@toot.community avatar

    @wogan Oh, that's a different variety of lighting, but yes, you are right about both home automation and DMX, it is quite the wasteland, and we also have Art-Net: https://en.wikipedia.org/wiki/Art-Net

    The trouble is that entertainment lighting is too small an area to attract enough programmers.

    I myself only do it for fun!!

    QasimRashid, to random
    @QasimRashid@mastodon.social avatar

    Calling it a gotcha that I "critique Biden policies after voting for him" is what's wrong with US politics today.

    Voting for him doesn't mean I can't critique him—it's literally the best reason to be critical. It's called holding electeds accountable. It's how democracy works. And it’s how political parties work when they aren’t cults.

    TomSwirly,
    @TomSwirly@toot.community avatar

    @AshleyMarineP @QasimRashid

    If there are two busses, both going away from where I need to go, it doesn't mean I'm "unable to negotiate" if I take neither.

    Nothing matters more than NOT killing our biosphere, yet D policies will kill our ecosystem just as fast as R policies.

    https://biologicaldiversity.org/w/news/press-releases/biden-administration-oil-gas-drilling-approvals-outpace-trumps-2023-01-24/

    https://archive.thinkprogress.org/obamas-worst-speech-ever-we-ve-added-enough-new-oil-and-gas-pipeline-to-encircle-the-earth-e5e24a156910/

    Watch D and R charge toward the Collapse for 50 years is horrifying. Watching people root for their teams as they kill our planet is nauseating.

    StillIRise1963, to random
    @StillIRise1963@mastodon.world avatar

    When fascists destroy something, it's an opportunity for us to build something better. Mastodon is the something better.

    TomSwirly,
    @TomSwirly@toot.community avatar

    @StillIRise1963 The early internet felt like this, and then it didn't: https://en.wikipedia.org/wiki/Eternal_September

    AOL was actually a good neighbor but we woke up and there were millions of new people, then overnight it was everyone, and the Fascists too.

    But I think "federated" is the key! Yes, Usenet was federated, but now we know, and we have fine-grained selectivity. I see in twenty years the Fediverse partitioned into two great "dark" and "bright" Mastodon cliques with very few grey instances.

  • All
  • Subscribed
  • Moderated
  • Favorites
  • JUstTest
  • kavyap
  • DreamBathrooms
  • thenastyranch
  • magazineikmin
  • InstantRegret
  • Durango
  • Youngstown
  • everett
  • slotface
  • rosin
  • cubers
  • mdbf
  • ngwrru68w68
  • anitta
  • GTA5RPClips
  • cisconetworking
  • osvaldo12
  • ethstaker
  • Leos
  • khanakhh
  • normalnudes
  • tester
  • modclub
  • tacticalgear
  • megavids
  • provamag3
  • lostlight
  • All magazines