←back to thread

Ask HN: LLM struggles to center div too?

4 points divan | 4 comments | 06 Sep 25 15:27 UTC | HN request time: 0.001s | source

I used AI coding agents for the past half a year a lot with Flutter. Recently I had to work on the tailwind web project, and I was surprised how often it gets layout/UI wrong. Even with puppeteer attached, haveing access to raw HTML output and screenshots, it's still struggles to do simple alignment, centering things.

In a way, it mirrors famouse "centering div in css" challenge, but I wonder if it's the same experience for other people who work with web stack more often.

1. alexander2002 ◴[06 Sep 25 19:40 UTC] No.45152269[source]▶
AI is really bad at fe.
replies(3): >>45152764 #>>45153208 #>>45154484 #
2. throwaway019254 ◴[06 Sep 25 20:56 UTC] No.45152764[source]▶
I had much better success with FE than with BE.

Does it mean AI is really bad at FE and even worse at BE?

3. dv_dt ◴[06 Sep 25 21:52 UTC] No.45153208[source]▶
Ai is ok at fe but to me seems best at raw html and ironically can really struggle with css and some frameworks and design systems

Im guess it has the most training samples on raw html followed by react, etc

4. muzani ◴[07 Sep 25 01:24 UTC] No.45154484[source]▶
Claude is exceptionally bad at FE; it has hardly improved since Claude 3 while everything else has made strides forward.

GPT seems legit multimodal, as in images and text in the same space, and it can maneuver FE at the level of an experienced engineer.