Do big tech companies train large language models on our personal data?

I don't see why they would, the average American's personal data is less coherent than the median written article on the internet. Unless you have evidence of it knowing much more than it should, it's doubtful your personal data is worth anything besides advertising entropy.