Inspired by this, we used multimodal learning to improve the performance ... the language models' ability to complete prediction tasks without task-specific labels. The self-supervised process ...
Results that may be inaccessible to you are currently showing.