The Future of Classification

contrastive loss is still going to be a key part, at least in the state of the art that I'm aware of. Classifiers also matter, for example, if maybe you have a lot of data that has a classification label and you want to use that in order to inform your learning. You can structure set contrastive problem as a classification one, although it's one option out of many, contrastives are really big problem. And of course, you can fine tune many of these embeddings in classifier type problems which we anticipate would be a massive use of these types of embeddings.