How is this possible if the average sample should produce average work?
Sample bias.
People who aren’t good at something don’t generally willingly engage in it. Or said alternatively, people who actively engage in topics or activities without being required to do so are the most likely to be highly skilled in those areas.
Taking that fact into consideration and that the internet is what fed ChatGPT it’s training data, it becomes clear that ChatGPT is likely to be better than a person chosen at random in a particular topic, however, at the moment it has several drawbacks:
1. Limited to mostly generalized topics.
2. Doesn’t know much about specific companies, people, etc.
3. Isn’t going to know much about topics that are highly technical niches mostly involving academic publications.
4. Not up-to-date. Data is from 2021 or earlier.
All that being said, ChatGPT can still be a useful ally.
None of this is Right 