Is That Different Than What You've Just Described?

The data set itself is noisy so often the caption just simply doesn't match the images an example you gave there. Images with better matching clip vectors are weighted more highly and that's what we can do with the static data set. We get results that are perceptually Strange to humans quite frequently relatively frequently, but to the model That the principle like the similarity principle doesn't apply The next thing we did was like look at what we could do with the like the live model itself And so we knew we would be getting these images After the fact so, you know, we couldn't look into the model while they were being generated But there are things you can do to basically recreate the generation process for