Yeh, I’m not sure if the likelihood of those images would change much if you omitted the word “kermit”. There’s a lot of talk in the AI community about the ability of DALL-E models to reason: many of these seem to just be combining concepts “frog” “insert movie title still” rather than doing reasoning about Kermit in these situations. Still impressive, considering where we were 5 years ago, but not quite what I think the OP wants to demonstrate.
3 Likes
