3 ms·
I don't understand how the title relates to the content of this article at all. They're even using CLIP which definitely has been trained.
by suddenlybananas 2y ago
I don't understand how the title relates to the content of this article at all. They're even using CLIP which definitely has been trained.
- dragonwriter 2y agoYou don't have to train the LLM soecifically for the tasks and even the auxiliary tools aren't trained on the tasks they are used as scorers for (because they aren't doing the task,just evaluating how well the LlM is), so there is no task-specific training.
- suddenlybananas 2y agoTask-specific training sure, but the title implies that vision itself is not trained.