In the context of this thread, "safety" refers to making sure we don't create an AGI that turns evil.
You're right that wrong answers are a problem, but plain old capitalism will sort that one out-- no one will want to pay $20/month for a chatbot that gets everything wrong.
>>Meekro+(OP)
How the thing can be called "AGI" if it has no concept of truth? Is it like "60% accuracy is not an AGI, but 65% is"? The argument can be made that 90% accuracy is worse than 60% (people will become more confident to trust the results blindly).