My AI Toolkit - Part 2
Special tools for special uses
Perplexity’s opening screen declares that it is “Where knowledge begins.” Perplexity describes itself as “an AI search engine designed to revolutionize the way you discover information.” The great thing about Perplexity is that it gives its sources so that you can verify its answers. Perplexity has several “Focus” modes (see below) that allow you to search the entire web or just academic sources, videos, or social media discussions. You can also solve math equations and generate text, but I haven’t used those modes much.
Perplexity is great at getting a quick start on academic research, but it will also help with things like finding useful videos. I like it for jump starting academic research because it actually cites its sources. The example below shows a quick search I did on trust in intelligent agents. Although I was disappointed that it didn’t cite my work from the 1990s, the results were solid. They’re kind of hard to see, but the little numbers at the end of its results are links to its sources.
In my experience, the sources aren’t the most definitive, but they’re good enough to get a quick handle on an area. In this case, the answer was actually quite good. In fact, if I needed to give a quick talk on trust in intelligent agents, I could easily base it on Perplexity’s answer while being confident that my presentation was accurate. By the way, Perplexity’s free version is so good I often forget to log into my Pro account. If you don’t have a Perplexity account, get one.
All of these tools I’ve discussed so far are great, and they all serve their purposes. But, if I could only have one AI service, I wouldn’t choose any of the above, I’d pick Poe.com.
Poe is all about versatility. It gives you access to dozens of large language models, including the big ones such the GPT, Claude, and Gemini models, along with image generators like DALL-E and Stable Diffusion. This is why I recommend Poe so often. It’s an easy way to try out the performance of the models used by the main AI chatbots.
Poe has some other nice features as well. It’s pretty easy to compare responses across models. Don’t like the answer Claude gives? Try a different model. An even more useful feature is the ability to switch between models within a single conversation, which is something the other tools can’t do. So, Poe isn’t the best at anything (in my experience); it’s superpower is flexibility. I routinely recommend Poe to my students because of this flexibility. Then, if they find themselves consistently using one model, ChatGPT for example, then switch to ChatGPT. Poe is simply a great way to learn about the nuances of the various models. It’s also handy to have if you’re favorite model goes down or you exceed your usage limits.
Well, there you have it, my guide to my very own, personal AI toolkit … except for one honorable mention and one intentional omission.
Honorable mention and intentional omission
Consensus earns an honorable mention because it offers some unique features. Consensus bills itself as “The leading GPT for research.” It’s a GPT because it’s based on OpenAI’s GPT models. In fact, it started life as ResearchGPT in OpenAI’s GPT marketplace. Consensus shows its DNA in its filters, which are shown below.
The filters reflect the way an academic researcher might search for journal articles and conference papers. It’s common for us to search by a year range, for example. The methods filter is interesting. It lets you search for meta analyses, systematic reviews, randomized trials and papers that used other specific methods. The Journals filter lets you limit searchers to those that meet a specific quality metric and Domains allows you to search papers in certain areas, such as business, computer science, and psychology.
Like Perplexity, Consensus cites its sources but it goes a step further. Consensus also gives very brief overviews of individual papers. You can also export the citations into a csv (comma separated values) or ris format. RIS is good for importing into reference management software like Zotero or EndNote.
If you ask a “yes or no” question, Consensus will give you the consensus of the research papers available. For example, if you ask whether creatine improves brain function, you get the following:
The consensus is pretty mixed on this one, as you can see in the Consensus Meter. In addition to the summary and Consensus Meter, Consensus shows the papers it found and the answer based on that paper (Yes, No, or Possibly). Although I don’t research many yes/no type questions, areas in which there is no consensus might be fruitful avenues for research projects, especially if you can figure out how to move towards a definitive answer. Consensus won’t give a consensus unless it can find at least five yes/no predictions in papers. I asked Consensus whether college student retention programs were successful and it couldn’t give a consensus. I’ll let you draw your own conclusions about this result.
Microsoft Copilot (and similar tools)
My intention omission is Microsoft Copilot and the like. To be clear, I’m talking about the Copilot that works within Office apps, not the Bing replacement (or rebranding). I want to like Copilot, I really do. It has the potential to be massively useful. But every single time I’ve tried it, I’ve been disappointed. And I even pay for a personal subscription to Copilot. I keep trying, but I just can’t find a good use for it. Sure, it has some handy features, but nothing I can’t do pretty easily with other tools. Maybe it’s just me though. My cohost on the AI Goes to College podcast (you are listening, aren’t you?) loves Copilot. So, if you have access to it, give it a try. Hopefully, you’ll have better luck than I will. (I’ll keep trying though; there’s just so much potential in the idea of Copilot.)
Concluding remarks
Well, that’s my AI toolkit. I use some of these every day, with few exceptions. Although I generally use the paid versions, I think all of these have free versions, except for Copilot for Office. The free versions are fine for most purposes, so don’t let money deter you from trying these amazing tools.
Which of these would I miss the most? That’s a question I ponder from time to time. It’s a toss-up, but I think the answer would be either ChatGPT or Claude. I love ChatGPT’s flexibility and voice interface, but Claude is amazing, especially with Artifacts. Fortunately, I won’t have to give any of these up any time soon.
What is your AI toolkit? Let me know by using the contact button at aigoestocollege.com or by emailing me at craig@AIGoesToCollege.com. Thanks for reading!!