ChatGPT is testing the native file analysis function, and DALL·E 3 can edit pictures!
On October 29, some users shared on social platforms that ChatGPT Plus is testing native file upload and analysis functions. It can ask questions and search uploaded PDF and other data files through text Q&A.
For example, upload a 50-page employee handbook PDF file and ask ChatGPT, can you help me summarize the 5 core points of the handbook? ; What penalties will new employees face if they break the rules?
In fact, the file analysis function is not launched recently. Users can achieve it with the help of third-party plug-in functions launched early by ChatGPT. However, it cannot compete with native in terms of performance, efficiency, and contextual understanding of massive data.
If the code parsing function previously launched by ChatGPT is added, users can directly generate visual charts based on uploaded files through question and answer methods, which is of great help to business personnel in finance, marketing, and marketing.
Without writing a single line of code or switching multiple tools, you can complete all data analysis and create visual data graphs on a single platform.
The sharing user demonstrated the PDF analysis function of ChatGPT: he uploaded a description document of the open source large language model Mistral 7B, and then asked questions to summarize the PDF document in two sentences.
ChatGPT answer: Mistral 7B is a 7 billion parameter large language model that outperforms existing models on various benchmarks, while using grouped query attention and sliding window attention to improve performance. It is released under the Apache 2.0 license and demonstrates high performance and efficient reasoning, especially in reasoning, mathematics and code generation tasks.
It also supports multiple rounds of in-depth questioning mode. You can continue to ask questions. How much higher is the test score of Mistral7B in the PIQA benchmark than Code llama?
Another important test function is that users can upload pictures to make photos in ChatGPT . This uses the newly released Vincentian graph model DALL·E 3 from ChatGPT, and demonstrates powerful semantic understanding capabilities.
For example, upload a picture of a golden retriever and then ask, let this dog run.
ChatGPT’s P picture function display
Upload an image of a capybara and have it recreated in Pixar style.
This kind of delicate semantic understanding and picture control function is simpler and more convenient than Midjourney. In addition, OpenAI has also optimized other functions.
For example, users don't have to select "Browse with Bing" mode from the GPT-4 drop-down menu, ChatGPT can contextually guess whether they need to be online.