3 ms·
Paper: https://arxiv.org/pdf/2501.18096 https://arxiv.org/pdf/2501.18096
by underdeserver 1y ago
Paper: https://arxiv.org/pdf/2501.18096 https://arxiv.org/pdf/2501.18096
- suddenlybananas 1y agoI 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 1y 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 1y agoTask-specific training sure, but the title implies that vision itself is not trained.