Content
@
0 reply
0 recast
0 reaction
Darryl Yeo đ ïž
@darrylyeo
Good take re: frontend tools used in AI-generated interfaces. I think React, Tailwind, Radix UI, shadcn, et al. are just a few of many possible âstepping stoneâ abstractions that in their current form are only used by AI because of abundant training examples and an arbitrary invariant of human maintainability. (Thought that is debatable now â I personally wouldnât want to maintain a codebase generated by v0.dev by hand in its current state.) Taken to its logical conclusion, a sufficiently powerful AI system in the future would know to optimize its way back down the human-invented ladder of abstraction (to HTML/CSS/JavaScript and maybe beyond) then build layers of the stack back up from its own first principles as and how it sees fit. https://x.com/tannerlinsley/status/1883209696229925282
5 replies
1 recast
11 reactions
matt đ
@matthewmorek
If you parametrize the query to LLM to include least invasive method of building something, itâll defer back to MDN and using native as much as possible. Iâve had epiphanies with HTML Dialog and Popover last week, and it begs a question of why are we looking for complex solution when simple ones already exist?
1 reply
1 recast
2 reactions
Darryl Yeo đ ïž
@darrylyeo
I guess itâs an awareness problem at this point; these browser features landed fairly recently. Likely people are already accustomed to their JavaScript-based implementations and see no need to change it, or theyâre two levels up the ladder of abstraction with a pre-made UI component library and at the mercy of upstream maintainers to adopt it.
0 reply
0 recast
1 reaction
