@glynmoody it's interesting that the author seems to think that human-like behaviour from models that were trained on human-authored content is unusual. I'd be more surprised if they developed identifiably non-human behaviors.