Exactly, and I think not every instance can be claimed to be a hallucination, there will be so much latent knowledge they might have explored.

It is likely we might see some AlphaGo type new styles in existing research workflows that AI might work out if there is some verification logic. Humans could probably never go into that space, or may be none of the researchers ever ventured there due to different reasons as progress in general is mostly always incremental.