LLM Training

Multilingual RLHF: The Hidden Quality Gap in LLM Training Data

Most LLMs are trained predominantly on English preference feedback. The quality gap this creates in non-English outputs is larger than most AI teams realise — and it's fixable.