Top AI models underperform in languages other than English

I hear a lot from people around me, Koreans, that LLMs are mostly underwhelming. These days I hear that less and less, but what I recommend is to use English.

A simple array of English words is often better than a full, well-written Korean sentence(s).

My personal and subjective experience is that English is usually better than Spanish (my native language). Unless I'm asking something specific about Spain such as laws, history, etc. I always use LLMs in English.

Most of my italian friends, prompt in Italian and they claim no degradation.

Are there some hard numbers on the latest models?

2 paragraph article estimates at 6 minutes read?

Am I not seeing most of the article?

On the content: speakers of a single language don’t realise this is the case, although they’ve always known that non-English countries tend to get worse service in general in software (like Siri voice, or speech recognition (and if you speak a minority language of your country, even worse), map services, or sometimes just the LLM answering with US customary units that you don’t understand nor expect, etc.)