Comment by embedding-shape

Wasn't that "twoot" (or whatever Mastodon calls them) made just a week ago? Unlikely to have been in the training dataset of a model becoming available for public use today, unless Google made some serious advancements on the training front.