holy shit this trash is token inefficient
>>105740256we have the llm write code using streamlit or similar for web interfaces
>>105740484its not a tokenizer issue, it's just verbose and shitty
>>105740530You said token inefficient, not character inefficient
There's no reason each tag and common syntax element cannot be its own token
>>105740572i'm talking about llm tokenizers.
>>105740591Tell me what you think a LLM tokenizer is
I just hate HTML because it's a pain in the ass to write all the closing tags. Never used a non-trivial amount of CSS so I can't really comment.
>>105740604tell me what YOU think they are and what they are buitl for
>>105740623You're not getting out that easily chuddie. You're the one making the claim that a LLM cannot be made to tokenize html efficiently, which is the unusual claim that requires proof here
>>105740636put it into a tokenizer and see how expensive it is you stupid cunt
having the llm roll its own html/css is usually 10-50x the token cost of having it use something like streamlit
>>105740710You're rage baiting, I don't think /g/ is quite this dumb
Feel free to refer back to my original point:
>>105740484
If you can't into CSS in big* 2025, unironically kys.
* :-)