Interesting point. For LLMs like GPT-3 & 4 we simply don't know because it's not disclosed. For BERT we do know though. There's separate BERT-English and BERT-multilingual models, for example, etc.

→ View original post on X — @rasbt,