Have you ever struggled with reading research papers? I have. In this guide, I’ll introduce a more effective method using Anthropic’s Claude and some AI prompt examples. With Claude, not only students, but anyone can benefit from the information research papers contain.
Why Use Claude (Free Version)
In this article, I’m focusing on using Claude 2 from Anthropic. It’s an artificial intelligence chatbot similar to ChatGPT but offers some distinct advantages with the free version. Namely, I can upload a PDF file. It also offers a larger token size which means it can handle more data than most systems. Like the free version of ChatGPT, you may hit a threshold where you have to wait before progressing.
The prompts and questions I’ve listed here are designed to get you started so you can build your own framework. While some of these questions could be considered universal, you’ll want to adjust based on the subject matter and goal. Also, most of the images are partial screen snaps to give an idea of what type of answers Claude returns.
To be fair, the paid version of ChatGPT can handle these tasks too. You can either upload the research paper or use a 3rd party plug-in called WebPilot.
Getting Started
Before submitting a research paper to Claude, I like to give it a first look. For example, I like to know:
- how is the paper laid out?
- how long is it?
- is the paper machine-readable?
- what’s the last section
Some research papers are nicely laid out and they are easy to read. However, I’ve seen some, where the PDF pages were images and the AI tools couldn’t extract the data. Two sources I go to for research papers are arVix and Google Scholar.
New to Claude 2?
Claude is a conversational AI tool and the “2” references the version or release number. The tool is available in both a free and paid plan. Once you create an account, you can converse with it in a question-and-answer format. Your questions are called “prompts”.
- [A] is a menu icon. It will open your previous sessions.
- [B] is the name of your current session. In this case, it reads “untitled” since we haven’t started.
- [C] is your profile icon. It links to your subscription plan and help.
- [D] is the chat area where you enter your prompt.
- [E] is an attachment icon where you can upload files such as PDFs.
- [F] is the button that submits your prompt.
Context is Crucial
Anytime you converse with an AI tool it helps to give it context. These tools will use that information when providing answers to your questions. In the example below, I’m reading a PDF research paper on protecting content management systems.
Unless I state my role, Claude doesn’t know how the information applies to me or why I’m interested. But by providing this context, I can help Claude AI frame a better answer for me. If I don’t provide this context and just submit the file, Claude will give me a summary.
Why Should I Read This Paper
If you look at my screen snap above, you’ll see I also asked Claude, “why is it essential I read this paper“? I’m trying to be efficient here. I borrowed the idea from Blinkist, which I previously reviewed. They have a book summary section called, “What’s in it for me?“
If Claude comes back and the answer doesn’t serve my needs, I’m probably going to find another research paper. Hopefully, you can see how adding my role shapes Claude’s response.
Based on the response, I want to continue my questioning.
Create a Section Highlights Table
This is a step I added for a couple of reasons. The first was I noticed sometimes these AI systems wouldn’t provide info on the whole document. This isn’t specific to Claude. That’s one reason I scan the PDF document before uploading. They might skip sections at the end. The second reason is I want to get a better idea of the terminology.
I also requested Claude to return the data in a table format. This is a personal preference and you may prefer not having tables or columns. Again, I’m trying to get the gist of the document.
Now, if my table comes up short based on my earlier read, I know I have an issue. In my case, I did get all the sections.
Ask for Help or Proof
Sometimes a prompt won’t work the way you want. In these cases, once I get the results I want, I ask the AI tool to help me rephrase the prompt. In the example below, I was having ChatGPT give me what I was looking for. That’s when I asked it for help.
You can use this approach to refine your prompts as well. For example, if you had a different exercise where you asked the tool to create 5 writing samples and there were 2 you liked. You could ask the tool, “Please refine my prompt to give me results like items #3 and #5.”
As good as these tools are, they sometimes make things up when they don’t have an answer for your prompt. If something isn’t correct or seems off, ask the tool. After all, if you were having a conversation with a friend and something seemed wrong, you would probably say, “Are you sure about that?”.
Another example is if you’re trying to compare findings from multiple research documents. Sometimes, you’re reading one study, but a finding in Paper A contradicts something in Paper B. Ask Claude to compare and contrast the findings.
Build a Glossary Table
I’ve never read a research paper where there were terms I didn’t fully understand. Sometimes, you can gather the meaning from the surrounding sentences. Other times, you can’t. Regardless, I find my comprehension increases if I review glossary terms.
Again, I’ve asked Claude AI to read the entire document and provide the Heading Reference for the term. And even though I previously told Claude I was a website publisher, I want the definition in non-technical terms.
Tell Me More About the Authors
I think it’s important to know about the paper’s authors, but I’m not fond of how most papers display them. Instead, I prefer to prompt the system for information.
Related Questions:
- Can you provide links to any other publications the authors may have written?
- What research methodology was used?
- How was this research funded?
Ask for Key Findings
Now that we have a good idea of the terminology and outline, I like to ask about key findings. More often than not, these items lead me to additional questions and clarification. For example, I wasn’t familiar with either WPScan or SCANTRAP.
Statistical Findings
Like key findings, stats are one of those items that people like to focus on. And sometimes they aren’t referenced in key findings so I like to pull them out. As you can see, some of the items aren’t statistical in nature but offer hints. For example, look at “most popular” and “fully”.
Setting Boundaries
Now this one relates back to why I read the PDF before starting my session. When doing a follow-up question, Claude presented some info I hadn’t seen when I scanned the initial doc. As you’ll see from my prompt, Claude did indicate it found those tools by doing its own research. This isn’t bad per se, as sometimes you want more info. You may recall from the earlier bio section, I asked what other papers the authors had published.
Based on your research, you may want to have Claude stick to just the document. All you need to do is tell Claude to stick with this PDF and not go elsewhere unless you explicitly provide instructions.
Any Actionable Steps
This is probably an optional prompt for many depending on the subject. In my case, I was looking for actions I could take. This time, I slightly altered my role and said I was the webmaster.
What Didn’t I Ask But Should’ve
This is one of my favorite prompts. It reminds me of the Rumsfeld Matrix and “unknowns” and “knowns”. This is where I often find things that weren’t on my radar.
