ChatGPT Tests Its Ability To Read And Interact With PDFs And Other File Uploads
Since its inception, ChatGPT has been an incredibly useful tool to generate ideas and content, write code, or perform more menial tasks that can free-up humans for higher level tasks. Even Boston Dynamics integrated the tool with its robotic dog Spot recently, to enable it to “talk.” Now, OpenAI is expanding ChatGPT’s capabilities by allowing file upload to allow the AI to read and interact with PDFs and other file types to generate responses.
ChatGPT Plus subscribers may notice a new feature rolling out that allows file upload directly in ChatGPT. According to reddit user, luokai, the model can chat about PDFs or other file types and generate a response based on information contained within them. We used this to upload a cover letter to get feedback on it, and while it is not perfect, as you can see below, it is certainly useful.
The other interesting thing that this could be used for is malware analysis, though when we tested it, there were some errors and warnings. ChatGPT says that “this environment is not a full-fledged malware analysis lab, so I'll be limited to static analysis and some other basic techniques.” Also, it did not like processing the samples we gave it generally, but we do not know if there is some file scanning happening on the backend or if it is a slightly buggy feature that is being improved.
Beyond the file reading, ChatGPT will also automatically select a model based on the conversation, so you do not have to choose what exactly you want to do going into a ChatGPT session. This will make interactions with ChatGPT significantly more seamless overall.
All in all, OpenAI’s new features for ChatGPT are impressive additions to the platform that make it stronger than ever. Hopefully, we will see these features roll out beyond just ChatGPT Plus members, but that may be quite a way down the line. In any event, if you want to test this out and you are a Plus subscriber, you can enable beta features in ChatGPT settings to test this and more.
Side note: the above hero image in this piece was generated by Dall-E 2 from OpenAI.