All LLMs are neural nets, not all neural nets are llms, but they're similar enough to have the same general flaws. 'Neural Networks" are misnomers, at best; especially given the designs were first being implemented before we had any real idea how neurons actually worked. It's why Brain Organoid interfaces still completely destroy entire simulated interfaces in pretty much any task we've managed to actually train them on.

It's also how we know we're not close to the software or hardware capability to actually do anything complex. The best that we've been able to do is simulate a fly's brain with a super computer.

Yes. That does seem to be the case based on the evidence before us.