I’m also on Mastodon as https://hachyderm.io/@BoydStephenSmithJr .

  • 0 Posts
  • 514 Comments
Joined 3 years ago
cake
Cake day: October 2nd, 2023

help-circle





  • We don’t even have standards that strong in programming languages or even fucking machine code (ISAs) anymore.

    I think I would like to return to that ideal time (if it ever existed), but… I feel like I’m in a vanishingly small minority.

    I think it comes down to incentive structure, and the most clear incentives push away from strong stnadards. The big advantage to (a) strong standard(s) is(are) interoperability, but that’s something end users have to demand because it’s an anathema to rent-seeking-behavior (a central facet of surveillance capitalism, choke-point capitalism, enshittification, and technofuedalism). But, even there, natural incentives fail us, since most users get more utility from “innovative” features instead of low switching costs – or at least the think they do until they actually try to exit a platform/service.










  • Sort of. A friend used it to generate some “tests” of questionable quality, a cousin is using it to help her learn and use a DSL (my term, not hers) for interactive tasks for her students, another friend was using it for source code generation, but I don’t recall the specific results.

    I disagree that it is no longer development, I see LLMs as yet another tool for generating code, and we’ve had generated “source” code since before C was standardized. I think the any code output by most LLMs is derivative of so many works under so many licenses that it is likely not possible to distribute it at all without violating some copyright and is certainly unacceptable for any Free Software project.; I think this is ethically true even if courts find LLM outputs are not derivative works or not subject to copyright protection at all – at least as long as copyright protects Disney. But, I know people that are working on a Free Software LLM, and “the Stack” provides enough information that you could provide all the necessary attributions for works derived from it.

    While LLM hallucinations are a real concern, they can be less impactful when doing code generation because of all the automated static checks plus the culture of peer-review. But, I also tend to favor languages with static type systems.