I'm curious if this is an artifact of using the Internet for training data. Whenever I see an extended argument on HN or Reddit, I rarely see anyone admit at some point that they think they're wrong rather than adamantly defending their initial statement.