A research team has demonstrated an innovative visual location recognition technology called Revisit Anything, which can identify the location in an image by simply inputting the picture. Image source note: The image is AI-generated, provided by the image licensing service Midjourney. This technology combines the latest SAM (Segment Anything Model) and DINO (Self-Distillation with No Labels) to enhance image segment retrieval.

To test the 'stress resistance' of these AIs, a group of researchers embarked on an 'adventure'. They created a test called Self-Contradictory Instructions (SCI), which is essentially the 'death challenge' for AI. This test includes 20,000 self-contradictory instructions, covering both language and visual domains. For example, you might be shown a picture of a cat and asked to describe this 'dog'. Isn't that a challenge for humans? Oh no, it's a challenge for AI.

During the pandemic, scanning products achieved growth against the trend. IDC reports that the shipment volume of scanners in China increased by 9.7% year-on-year in 2021. Google's Tesseract OCR engine boasts an almost 100% recognition accuracy, greatly addressing data accuracy issues. Quark Scan King has launched a万能 scanning function that supports editing of text and table content, enhancing user experience. Quark Scan King addresses young users' pain points with automatic document format and layout recognition. Large model technology supports Quark Scan King's handwriting and formula recognition accuracy reaching 99%, leading the industry.

🎯 The Trend of Large Model Applications: The article mentions that large models have evolved from model-level development to innovative application-level advancements. 👀 Limitations of Visual Recognition: GPT-4 demonstrates limitations in visual recognition tasks, highlighting the need to improve the model's generalization ability. 💡 The Importance of Model Testing: It is crucial to be cautious of only testing the model on the training set, as generalization performance is key.