While someone doesn't seem to like the point, the point that a few inputs (i.e. people) are unlikely to have a material difference on the outputs of a model remains.

If someone wants to make a copyright or privacy argument, they're free to, but I'm wary of making exaggerations here when that could lead to unintended consequences.

For the record, OpenAI is specifically worse with that, although it is also more restrictive. It's hard to say why. It could be that whatever architecture they use to appear more sophisticated is more prone to this.