This is definitely part of the issue, not sure why people are downvoting this. That's also why tests like this are important, to illustrate that thinking in the way we know it isn't happening in these models.