In this article, we'll explore BLIP-2, a new Vision Language Model by Salesforce which beats the previous state-of-the-art Flamingo on various multimodal tasks.