Analyze an archive of PDFs - ChatGPT Tutorial

- [Instructor] One of the biggest strengths of custom GPTs is that you can upload and use your own content or data. Rather than creating a tool solely around web searching or content that's available elsewhere, if you have files or documents specific to your work or interests, you can create a GPT assistant to help you manage, interpret, generate reports, or do just about anything else you can imagine doing with your files or data. For example, here on my desktop, I have a folder of a year's worth of old newsletters in PDF format and I want to create a GPT to help me figure out what topics and information are covered in them and to be able to pull out specific information from this archive as needed. So we'll go back to ChatGPT through the web interface. Currently, you can't create custom GPTs through the desktop or mobile apps. And here we'll click Explore GPTs on the sidebar. That brings us to the GPTs page. And here we can browse GPTs that other people have created and uploaded to the GPT Store. But in this case, we want to create our own, so we'll click Create. And from here we can create our GPTs in one of two ways, or in a combination of both ways. We have the Configure tab where we can name our GPT, add a description, and provide real language instructions on what it should do, how it should be behave, and so on. But the easier way to create a GPT, at least to get started, is to come back to the Create tab where you can build your GPT using just conversational language and letting the GPT builder ask you questions to refine your GPT. So, for example, down here on the left of the panel, I'll tell it, "I want to create an assistant to analyze, report on, and extract the contents of a collection of PDF newsletters that I'll upload." So we'll send that. It tells me it's come up with a name for this assistant, suggesting Newsletter Analyst. You can call it whatever you like. I'll just say "Keep that name." And next it's going to generate an image for our GPT. Again, I'm fine with that so I'll say "Keep it." Now after the name and image, you'll usually be asked some questions to help you refine what the assistant does, the tone it uses, the content it will focus on and so on. So you can use the prompts or questions it asks you, or you can just continue typing instructions on your own. Maybe in this case I'll say, "This assistant should focus on extracting key information, summarizing content, and identifying specific topics and trends across all the documents." So the GPT builder has updated the information saying it's going to focus on that information and is asking, "Next, how do you want it to talk? Should it be formal, friendly, technical, casual and so on?" So we'll continue answering its questions. I'll say, "Keep the tone conversational. Provide factual analysis and avoid personal opinions. To ensure accuracy, ask the user for clarification in situations with unclear requests." Basically, you want to put yourself in the mindset of how you'll really use this tool and try to imagine the places where the GPT might not know what to do and try to provide instructions to cover those situations. Now since I'm working with my own documents, I need to upload them so the assistant can analyze them. So I'll click the plus button here to browse for my files. So you can see we can upload from Google Drive or Microsoft OneDrive if our files are online. But in this case, the files are on my computer, so I'll click upload from computer, and we'll browse out to the desktop. As of this recording, the largest number of files you can upload at once is 10. So I'm just going to select the first batch of files here. I'll choose open, and then we'll upload those. And while that's happening, we'll upload more files, and we'll just grab the second batch here. Now, it's currently analyzing the first batch. I'll just let it finish. And now upload the second batch. And now it's just giving me this analysis of the second batch. I don't really need to read through all this right now. But let's go over to the Configure tab again. So if you want, you can update or change the name or description of the GPT from here, and you can read or edit the instructions that we've generated. So you're free to add or change the instructions from here, but you can also switch back to the Create tab at any time to do it conversationally. Notice we also have some conversation starters, which it suggested, and we can edit or delete or add to them as well. And we can see the conversation starters as they'll appear in the GPT here in the preview area. Now below that, we can see the files we've uploaded as knowledge. If we need to add more newsletters later, I can come in here and click upload files. And below that we can choose which of these capabilities we'll allow it to have. For example, if I want to strictly limit its knowledge to just the newsletters, I can uncheck web search. And in this case, I won't need to generate any graphics or presentation materials so I can uncheck Canvas. And I probably also don't need it to generate images with DALL-E or to use the Code Interpreter, so we'll just uncheck that as well. So with these items unchecked, the knowledge will come primarily from the files we've uploaded. All right, so let's ask it some questions over here in the preview panel to test it out. So I might ask something like, "What is a commonly recurring topic throughout all of the newsletters?" So it takes a moment, it looks through all the newsletters and gives me the summary. Or I might ask something like, "What were some of the web design related topics mentioned?" And, again, we get a thorough summary, and it even asks me if I would like a list of the recurring tools or events in a downloadable format. So now I have this tool I can pull up at any time to ask questions about the content of my specific files. This is so much faster and useful than just using a simple find command in a PDF reader. And the possibilities are endless. You could upload the user manual for your car. So you can ask conversational questions about how to program the stereo instead of looking it up. Or upload your company's HR manual, so you can ask if decisions you're thinking of making are in compliance with the company's policies. Now, if we need to make any modifications, we can switch back to the builder side and add some more instructions. I'll go back to the Create tab. And from here, we can always just add more instructions by typing them in here at the bottom. So I might say "Keep the summaries concise." Maybe I'm thinking these summaries are a little too long, but to "Provide more detail when asked." And once you're happy, you can click Create. And for this example, I'm going to save this GPT just for me. But if these newsletters contained info that my colleagues or others might want to be able to look into, I could choose anyone with the link or even make it public by putting it on the GPT store. But in this case, again, we'll just keep only me selected and click Save. And here I can click View GPT. And now we're looking at the GPT. And I can pull this up at any time by going to Explore GPT to My GPTs. So anytime I need to look into the contents of my newsletter archive, the GPT is ready for me. And if need be, I can come in and choose Edit GPT to continue refining it.