[anon_cow1111]

I'm having a hard time believing this site is honest, especially with how ridiculous the scaling and rotation of numbers is for most of them. I dumped his prompt into chatgpt to try it myself and it did create a very neat clock face with the numbers at the correct position+animated second hand, it just got the exact time wrong, being a few hours off.

Edit: the time may actually have been perfect now that I account for my isp's geo-located time zone

[Zopieux]

On the contrary, in my experience this is very typical of the average failure mode / output of early 2025 LLMs for HTML of SVG.

[perfmode]

i read that the OP limited the output to 2000 tokens.

[lanewinfield]

^ this! there's a lot of clocks to generate so I've challenged it to stick to a small(er) amount of code

[fouc]

I wonder if you would get better results if you tell the LLM there's a token limit in the prompt.

something like "You only have 1000 tokens. Generate an analog clock showing ${time}, with a CSS animated second hand. Make it responsive and use a white background. Return ONLY the HTML/CSS code with no markdown formatting"

[anon_cow1111]

I got a ~1600 character reply from gpt, including spaces and it worked first shot dumping into an html doc. I think that probably fits ok in the limit? (If I missed something obvious feel free to tell me I'm an idiot)

[Springtime]

On the second minute I had the AI World Clocks site open the GPT-5 generated version displayed a perfect clock. Its clock before and every clock from it since has had very apparent issues though.

If you could get a perfect clock several times for the identical prompt in fresh contexts with the same model then it'd be a better comparison. Potentially the ChatGPT site you're using though is doing some adjustments that the API fed version isn't.