Training. Its just the awareness of the collective users of X showing up as Grok responses.

Basically, we can’t verify everything that AI says. Verification is still a human labour.