Why I'm using Claude more than ChatGPT
Let me start by saying that I still love ChatGPT and use it quite a bit. But Claude ... oh my, Claude has gotten good, REALLY good. The responses are excellent, Claude Haiku 3 (the stripped down model) is the best combination of response quality and speed that I've seen. Claude Sonnet 3.5 is as good as it gets in terms of response quality. Haiku isn't far behind though, and it's noticeably faster.
These alone aren't enough to make me use Claude more than ChatGPT though. The differences between the two aren't enough to make me fool with using two different models. The tipping point for me is the artifacts feature, which was added to Claude not too long ago. For the right use, artifacts makes a HUGE difference.
It's a pretty simple concept. When Claude creates something, the creation appears in a separate area on the right of the window. The conversation is still visible on the left. Check out this example:
I asked Claude for a list of higher ed use cases for the artifacts feature. You can see the response on the left and the list on the right. Now, I can simply copy the list and paste it into whatever I'm working on, or I can download the list as a markdown format file. (A markdown file uses special characters to format a document. It sounds complicated but really isn't. You can learn more about markdown here: https://www.markdownguide.org/getting-started/).
The artifact can also be published to the web, which might be useful in some cases. At least that's the theory. I've had mixed luck with publishing artifacts, so I'm hesitant to recommend this particular feature. Publishing could be handy when you want to quickly share an artifact, but I don't seem to use this feature much. I can see where it might be useful though.
Let me give you a real-life example of how I recently used Claude's artifacts. I was working on a contract job to create some learning materials for a governmental entity. The project was already well underway by the time I joined, so I was playing catch-up and had to create the materials quickly, but with high quality. Feedback from an earlier phase indicated that the examples used were too generic; the client wanted more relevant datasets and learning activities. Since I knew virtually nothing about the context (always fun!), I turned to Claude for help. Claude was able to give me a quick overview of the context. This was handy, but I could have gotten the same info through a couple of phone calls. Where Claude really shined was creating the relevant activities and datasets.
One of my topics was Excel (at a fairly advanced level). I've taught Excel a lot, so I had plenty of datasets and learning activities but they were all fairly generic business examples. I was able to give Claude some parameters in a straightforward prompt and Claude cranked out some good activities and (perhaps more importantly) the datasets to go along with the activities. Searching for specific datasets online is a bit of a nightmare and a definite time sink. Creating datasets is just as bad.
In a matter of 15 minutes or so, Claude was able to help me create a good activity and dataset. I did have to do a little tweaking, but Claude got me most of the way there. For example, Claude created a Name column with given names and surnames. I wanted them separate and it was easier to just do this with an Excel function. Even with the modifications, I was able to create the activities in a fraction of the time it would have taken without AI. The artifacts feature was especially useful in this context. I could quickly see the dataset, which made it really easy to see what modifications I needed. You kind of need to try it for yourself to see what I mean, but trust me, artifacts is a huge plus in the right situation.
Here's an example. The prompt is pretty simple; I just described what I was working on and what I wanted. (It turns out I should have been a little more specific, which I'll get into in a minute.) The prompt is intentionally a little vague since I wanted to see what Claude would come up with. If I had something more specific in mind (for example, specific columns), I would have added that information to the prompt.
Can you see what I should have included in the prompt? Claude only created 10 rows of data. That's easy to fix, just tell Claude to modify the dataset.
Claude has a tendency to omit much of the data. You can see where it says that it omitted 80 rows for brevity. That's fine if you're still adjusting things, but once you're happy with the dataset you'll want all of the data. Just ask Claude to produce the entire dataset. Sometimes it takes a few tries. If you want larger datasets, you might want to use Claude 3.5 Sonnet rather than Haiku. In fact, I recommend using Sonnet for creating datasets generally. It just does a better job and provides more information, such as descriptions of the columns in the dataset. One warning: If you want to create long datasets you may need to do it in stages if you run into response length limitations. This is a little inconvenient, but still MUCH better than trying to track down datasets.
Also, you can scroll back to earlier parts of the conversation and click where it says "Click to open code" and you can see the earlier artifact. Once you have what you want, just click on the clipboard icon to copy the data or the download icon to download the file. In this case, it will be a csv (comma separated values) file, which is easy to bring into Excel.
If you wanted to take this a set further, you could easily ask Claude to generate a number of decisions or questions that can be addressed using pivot tables and the dataset. You can also ask Claude to generate detailed instructions for labs or assignments using the dataset. Keep in mind that you will need to double-check the instructions before giving them to students.
Before closing, I need to mention some limitations of Claude. Here's a list of some things that Claude CAN'T do:
Create images
Create custom GPTs
Remember things across conversations
Access the Internet in real-time
Use plugins
So, there are still some reasons to use ChatGPT over Claude. But for tasks such as creating class activities and related resources, Claude is my choice (for now). What AI tools do you use the most? Let me know: craig@AIGoesToCollege.com.
Remember to listen to the AI Goes to College podcast, which I co-host with Dr. Robert E. Crossler (better known as Rob) from Washington State University. Rob and I have some great conversations about how we're using AI and how we see AI changing higher ed. Check it out at https://www.aigoestocollege.com/ or wherever you listen to podcasts.