DeepSeek has introduced its latest model, DeepSeek-OCR 2, which employs the innovative DeepEncoder V2 method. According to PANews, this approach allows artificial intelligence to dynamically rearrange image components based on their meaning, rather than merely scanning from left to right. This technique mimics the logical process humans use when observing scenes. The model demonstrates superior performance in handling complex images, such as documents or charts, compared to traditional visual-language models, achieving more intelligent and causally-inferential visual understanding.

#AI