> But AI could be made stricter and not hallucinate

Do you have a source for this? So far every single LLM-like does a lot of hallucinating, often just small things like wording that implies stronger correlations than exist, or wording that implies relationships that don't exist, etc. but sometimes it's also just wrong facts.