After the emergence of ChatGPT, developers around the world have developed countless AI tools based on it. Today, we will introduce three assistants that can converse with documents, and I will provide detailed information about their differences and usage.
First, the user-friendly PDFchat:
PDFchat is a document GPT project that requires the use of its own API. Its GitHub project website is very simple: https://github.com/bhaskatripathi/pdfGPT. It supports local deployment and also has an online webpage. Here, I will demonstrate using the web version. Just fill in your API key on the left and enter the document URL or upload a document and input your own question below to get an answer on the right. ChatGPT will automatically learn the document content and provide answers. I uploaded two documents, "Bitcoin Whitepaper" and Google's "Transformer" paper. Here are its answers. The downside is that the web version cannot save content and the page is not aesthetically pleasing.
Second, the powerful chatpdf:
Chatpdf is an online tool available at https://www.chatpdf.com/. It can be used for free without entering an API, and the page is aesthetically pleasing. It also provides examples. You can upload a document and use it to ask questions. After logging in, you can also save the conversation. However, please note that it can only upload PDF documents within 120 pages for free. Beyond that limit, it becomes a paid service and not cheap.
Third, the domestic open-source literature assistant chatpaper:
This tool was developed by a Chinese developer and is completely open-source. It can be deployed locally to ensure security. The GitHub link is: https://github.com/kaixindelele/ChatPaper. Here, I will only demonstrate the web version, which is similar to chatpdf and has a user-friendly interface: https://chatpaper.org/. Just upload the document, and it will list the original document on the left and provide an overall summary. This is very suitable for researchers who want to quickly find the literature they need. Additionally, each answer also includes the relevant page numbers, making it a great tool for reading literature. However, the current number of free tokens is limited. I asked one question and already used 2/5 tokens. Users can choose to pay or use their own API based on their needs.
Finally, it is important to note that all AI-based document reading tools provide answers for reference only and may not always be correct. Their understanding capabilities are still limited.