Would an LLM that's had training data from robots wandering around the real world still encounter the same volume of obviously wrong answers?

Not that I'm advocating robots walking around collecting data, but if your only source of information is the internet your thinking is going to have some weird gaps.