Mathilde and team have worked very hard on an important child safety AI benchmark Mistral massively improved their score (a great thing) and KORA bench was cited on their model card It's rare to see a benchmark have direct model improvements like this. Very cool