It’s even more than just “accepting everything as true” the machines have no concept of true. The machine doesn’t think. It’s a combination of three processes: prediction algorithm for the next word, algorithm that compares grammar and sentence structure parity, and at least one algorithm to help police the other two for problematic statements.

Clearly the problem is with that last step, but the solution would be a human or a general intelligience, meaning the current models in use will never progress beyond this point.