That you can’t assume everything is possible just because some impressive things have happened. These breakthroughs didn’t come out of nowhere, they’re based in very sound and rigorous theories. You have provided none as to how or why a language model would be able to logically verify what it produces. You’re just vaguely gesturing at the fact breakthroughs exist.