Here is some more information on the problem: https://news.ycombinator.com/item?id=11361398 https://news.ycombinator.com/item?id=11361398 Apparently some of th

Here is some more information on the problem: https://news.ycombinator.com/item?id=11361398 https://news.ycombinator.com/item?id=11361398 Apparently some of the questionable responses are partly identical to months old tweets, so it seems in addition to some parroting 'vulnerability' (which was exploited by 4chan) they also had a poorly sanitized training set to begin with. It seems odd that this was not mentioned in this public statement.