You can prove that most people don't learn CSS using most LLMs: they're trained on github code.

A card with a top image, some text and a button - which should be...

- A card

- A top image

- Some text

- A button

ie 4 HTML elements, ends up being about 10 HTML elements with various strange hacks in the div-soup HTML the models have scraped from Github.

Then someone else comes along and uses tailwind, because naming 10 arbitrary HTML elements (rather than .card, .card img, .card p, .card button or similar) is hard. They're right, but the problem is they didn't need that many elements in the first place and wouldn't if they'd just learnt CSS.