Posted by rdslw 1 day ago
So I asked my agent to write typst, ran "typst watch", and now I can look at a nice pdf file. it even auto-refreshes when the clanker changes it.
Edit: Sure there is some small overlap here, but it's really not comparable and definitely not like the way the author describes things. User personalization in Emacs has normally been on a much smaller scale than rewriting entire packages. Configuration is generally smaller tweaks or things on top of existing packages because Emacs provides cohesive extensibility to the point that it often doesn't require "rolling your own." Most packages are already extremely configurable and tailorable. You don't magically get that sort of environment with LLMs. Emacs is much more cooperative/generalized.
The scale and type of custom/personalized software we're seeing now with AI is completely different from how things have been in Emacs. I'm not saying that's a good or bad thing (I think it's both), but it's very different from Emacs and definitely more comparable to something like vim/neovim where (in part just because of the sheer popularity) you constantly have people "rolling their own" packages and a billion versions of everything. Even that is not a great analogy. This is something completely new.
Specifically the idea that people generally just ignore existing versions of packages and make their own has never been the case, especially compared to other editors (even VSCode).
> There are popular elisp packages lots of people use. But except for Magit, nerds are alarmingly apt to replace them with their own shinier versions (and then to show them off, transitioning to the spore-forming phase of the elisp lifecycle). Everything in Emacs is malleable.
> Until now, the Achilles heel of Emacs culture has been that, except for Magit, its packages tend to be wretched user experiences. Ugly, slow, and discoverable only after inflicting years of elisp cortical injuries on yourself.
If this is the starting thought, I don't know how you wrap back around to publishing and advertising the generated code.
Either you create the best possible mac markdown viewer and should share it as that, orthogonal to any statement of AI use. Or you're just adding to the noise of tools available online. Where other people should ignore your work, and go slopcode their own markdown viewer.
"With LLMs you can just make your own".
I read dogleash's comment as emphasising the bad parts of that. If everyone's just sharing their half-baked slop, it becomes harder for people to discover the good programs which are worth using, and harder for those quality programs to get social proof (or good direction).
"Content creation for an audience of one" is really the revolutionary change that is happening right now because of AI. Disposable apps, disposable books, disposable movies, disposable music. Things that are made for a single person, used once or a handful of times and then thrown away. The entire economic model of content creation and distribution is going to explode in the next 3 or 4 years, and very few people are prepared for it.
> The entire economic model of content creation and distribution is going to explode in the next 3 or 4 years
I think this is also inherently self-contradictory. What's the point of distributing content made for one? This gets into the same fallacy that people engage in w.r.t. "applications for one" displacing software developers. Yes, LLMs can pump out buggy software that suits one person's needs, and it doesn't need to be reliable enough to deploy at scale. It serves real utility here, because there was a gap between "the value of such software" and "what software developers are willing to work for", which meant that this software wasn't being created because there wasn't economic value in it. But then, how does one suppose software that has no economic value is going to replace all the professional software developers who were being paid to produce software that has economic value? LLMs filled a gap software developers weren't being paid to do, but given that they were not paid to do it, their jobs are not contingent on the existence of this niche. It simply doesn't follow that being able to produce content with zero economic value, whether that's applications or content for one, will cause an 'explosion' in the existing economic models.
Applications for one, might serve as a sort of "ultra-processed software" that can fill useful niches like generating textbooks on the fly (https://news.ycombinator.com/item?id=48130679), generating focus oriented music for long study sessions, and obviate writer's block.
However, I feel like there are downsides that are both obvious (insularity, discoverability, reliability, and platform dependance) and non-obvious issues which will take time for the public to determine what they are.