Если вкратце, то он понимает, что изображено на картинках.
Видео презентации:
Click to view
Статья, хоть и без технических подробностей:
GPT-4 Technical Report"We report the development of GPT-4, a large-scale, multimodal model which can accept image and text inputs and produce text outputs. While less capable than humans in many real-world scenarios, GPT-4
(
Read more... )