Results with LLMs seem so random! Hard to say how long a certain task will take.

Trying to vibe code a better website (in SvelteKit and Tailwind) for Zapstore.

Started with Gemini 2.5 Pro (apparently best model ever for coding), it generated a broken site and was going in circles trying to fix. o4 mini currently struggling too.

On the other hand Gemini ported a complex niche Rust library to Dart in two prompts, and I was blown away.

Reply to this note

Please Login to reply.

Discussion

grok 3 is giga for coding, with the "think" mode enabled. 3.5 should be next week and supposedly even better

I'm a hobbyist, but LLMs seem to really struggle with frontend work; anything that requires design sensibilities. Which figures, I guess.

Possible! I had good results with websites before tho. The amount of Javascript and CSS out there for training is bigger than anything else

What's your go-to for CSS?

I asked for Tailwind

True story :110percent: