Spaces:
Running
knowledge and terminology
I assume that you are trying to get creative writing from models. But the fact is that Mistral and many others himself, despite supporting multilingualism, does not seem to be familiar with Wikipedia. You can endlessly improve your thinking abilities, but without expanding your knowledge, it won't have a breakthrough effect. To be honest, I tried to use the Gemma, Mistral, Command-R and other LLMs models to translate subtitles for a popular science film about history and nature. Almost everything is fine, except that their models don't know the real history and zoology, historical sites, nomenclature of species, and God knows what else.
In addition, many even top ranking models display the final subtitles in solid text or breaking the original structure, without numbering, line by line, and so on, and not as described in the assignment.
Since the creators of the LLMs himself does not seem to be very concerned about this problem, I would ask you to pump their brains with scientific dictionaries and Wikipedia. This is also necessary along with the support of other languages in which terminology must be adapted in a general scientific context, as well as idioms in a cultural context. I am currently unable to find an alternative to Gemini and Claude among the entire HF repository.
Thanks
You're right, current LLMs like mistral and others struggle with specialized knowledge like history and scientific accuracy, and often mess up formatting; it seems like giving them extra data like scientific dictionaries and wiki would really help them get better and I agree that's a big issue; we will look into that in our next iterations!