There’s already more than enough training data out there. The important thing that remains is to filter it so it doesn’t also include humanity’s stupidest data.
That and make the algorithms smarter so they are resistant to hallucination and misinformation - that’s not a data problem, it’s an architecture problem.