My AI Toolkit: Part 1
What AI tools do I use?
“What AI tools do you use?” That’s a question I get asked all the time, and I totally get the confusion. With so many chatbots, AI models, and AI assistants out there, choosing the right one for the right task can be overwhelming. People sometimes ask me this question; it’s happened frequently enough to cause me to write this article. What’s the best AI chatbot? Well, that depends on what you’re trying to do. The trick is to match the tool to the task. In this guide, I’ll pull back the curtain on my personal AI toolkit. This guide focuses on chatbots, but I’ll write a future article about more specialized tools. The order in which I discuss the tools roughly matches the frequency with which I use them. The complete guide was a little long, so I’m breaking it into two parts. In this part, I talk about Gemini, ChatGPT and Claude. Part 2 will discuss some more specialized tools, including the one I would choose if I could only have one AI service. Cliffhanger!
Google’s AI chatbot is insanely useful for getting answers to a variety of quick questions, especially if it can use its integration with other Google apps such as Maps or YouTube. (I wrote an article about this recently ADD LINK.) Here’s a partial list of my recent chats: How to make an upside-down exclamation (¡) point on my Mac (it’s Option-1), enabling markdown language in Google Docs, fixing an iPhone calendar sync problem, shoe glue versus boot glue (they’re the same), troubleshooting an Excel formula, how frequently I should check the water in my new golf cart (well, new to me), and which songs Paul McCartney wrote without John Lennon. Oh, I almost forgot. I also used Gemini to make sure I was using the correct Spanish phrase for an anniversary message to my wife, Tracy. It’s hard not to admire Gemini’s flexibility!
Gemini is great at these kinds of focused question. It’s capable of much more, but I find it much better than Claude or ChatGPT for the kinds of questions that are typically answered through a web search. I use Gemini so frequently that it’s almost always open on a secondary screen. When it’s appropriate, Gemini will also provide links to related websites, which can be useful. The next time you have a question, instead of bringing doing a search, try asking Gemini. You won’t be disappointed.
As much as I like Gemini, it’s not the greatest when it comes to helping me co-create things like this article. For those tasks, I turn to either Claude.ai or ChatGPT, depending on the exact nature of the task. I’m not sure which of these I use more frequently, so I’m covering them in alphabetical order.
ChatGPT is my general-purpose go-to AI chatbot. It’s the one that brought generative AI into the public light, and it’s the first one I used. The Mac and iOS apps are really good. In fact, the ChatGPT app is usually running on my Mac. If I want to have a conversation with an AI chatbot, I almost always use ChatGPT or Claude. ChatGPT is more versatile than Claude in some ways. For example ChatGPT can create images, which is something Claude can’t do. ChatGPT’s GPT-4o model is also fast, which is nice. (GPT-4o mini is even faster.) OpenAI recently added two new models o1 and o1-mini, which are supposed to be much better at reasoning. I’m still testing these, but they show promise.
I really like how ChatGPT and Claude can help me refine my thoughts and figure things out. A couple of days ago, I was thinking about the future of AI agent networks, which is a bunch of focused AI agents working together to accomplish complex tasks. I was trying to think through the barriers to the widespread use of these networks and when we might expect to see them used by average folks. These are complex topics and ChatGPT was great at helping me refine my thoughts. Given my work on this newsletter and as an academic researcher, I think through these sorts of things virtually every day. Pre-ChatGPT, I would either try to figure things out by myself or wait until I could talk through them with a colleague. Now, I have 24/7 access to a virtual colleague. Is it as good as talking with a colleague? Probably not, but availability and patience have their advantages.
Recently, I’ve found myself using ChatGPT’s voice interface more and more. I just got access to the new voice interface on the iOS app and it’s amazing. The slightly too long pause that preceded responses is gone. With the update, conversations seem about 90% natural … you can even interrupt ChatGPT, which is awesome (and oddly rude).
Claude is amazing in many ways. It’s responses are consistently good, and with the exception of a conversation voice interface and image creation capabilities, pretty much anything you can do with ChatGPT, you can do with Claude. So, much of what I wrote about ChatGPT applies to Claude as well.
Claude shines when I want to create something. Here’s a recent example. I had to create an assessment plan for a new graduate certificate program. These plans are a pain to create. Often it takes several hours of creating things like assessment methodologies and rubrics. With Claude, in under 30 minutes I had a very solid assessment plan. Claude’s Artifacts is what makes Claude so much better than ChatGPT for this task. I wrote about this earlier LINK, but here’s the basic idea. When Claude puts the creation in a separate window on the right, like this:
This makes it really easy to see what refinements the artifact might need. For example, Claude originally put four levels of competency in the rubric. Three was enough, so I asked it to simplify the rubric to only three levels and Claude responded. This might not sound like a big deal, but it is. Just being able to see the artifact and the chat at the same time makes refinement much easier.
I also switch back-and-forth between Claude and ChatGPT when I’m not getting good responses from one or the other. This seems to be a little random. (At least I haven’t figured out a pattern yet.) Claude seems a little more conversational to me sometimes as well. It will often ask follow-up questions, which is often useful and sometimes annoying, just like a real-life conversation.
One ongoing problem with large language models and AI chatbots is that they will make stuff up. The technical term for this is confabulation (which is a fun word), although some people call this hallucinating. The “made up stuff” sounds plausible, but isn’t correct. Sometimes this doesn’t matter too much, but other times you need to be sure. When being absolutely right matters, I turn to Perplexity.ai, which will actually cite its sources. To learn more, be sure to tune in to Part 2!