I’m interested in the idea of students using auto-generated content to create their own personal pathways through a curriculum of information, so have been looking at a couple of tools that enable this.
A few days ago I wrote about using STORM to generate a Wikipedia-like outline in response to the question: “What is the most likely impact of generative AI on the higher education system, taken to it’s logical end-point?”
Today I wanted to experiment a bit with using Perplexity Pages to automatically generate Wikipedia-like articles, using the same question.
The first attempt wasn’t exactly what I was expecting and it took me a few minutes of reflection to understand why. To some extent, we’ve been trained to ask generative AI a question (or give it an instruction), and it goes from there. My experience with STORM was like this; you just give it the starting prompt and it creates everything, with very little for you to do.
This was a different experience entirely, and not in a bad way. I just don’t think Perplexity Pages is suitable for answering questions or starting with a single prompt. I think it works best when you approach it with a pre-determined structure or outline in mind, and then use Pages to expand that structure into a larger piece of work.
In other words, the use-case and methods of using Perplexity Pages is different to when you use other frontier models. For example, I use Claude for both use-cases (i.e. for developing outlines and expanding on those outlines).
So anyway, I gave it the initial prompt and then sat back, expecting it to create something STORM-like. And then realised that I’d actually need to do some work to get it into a usable form.
Perplexity Pages is more customisable than I anticipated, enabling you to edit the original question, add sections, move sections around, add media, and so on.
And I like that you can specify some basic detail about what the new section will include (e.g. text-only, text + image, image-only), as well as some options around the depth of the information it generates.
Initially the “Add Media” option inserts an image that Perplexity thinks is suitable. You can also choose additional options, upload your own image, or generate an image based on the text in the section (you need the paid version for this). I also like the section summary that it provides.
In the example above, Perplexity automatically added an image that I didn’t think was relevant, although I could have taken a few minutes to explore the source material and it may have made more sense. You may find that, depending on how the information in the section is structured, your automated image / media selections are better than mine were.
I like the way that the information sources are easily accessible for each section, but tucked away to the side when you expand them. I’d like the option to start a section with my own sources but I didn’t see anything like that. If it’s not available, it feels like a missed opportunity to address the concerns around the accuracy of the final article. But maybe I just missed seeing it.
Even though the initial experience was a bit discombobulating, I found that once I’d wrapped my head around how to use Pages more intentionally, I quite enjoyed the process. I didn’t notice anything obviously wrong in the generated content, but if I’m honest, I wasn’t looking at it very closely. For this kind of high-level outline of a topic, I’d be very surprised if Perplexity (or any frontier model) was creating anything obviously wrong.
This was a useful experiment to learn more about Perplexity Pages, and I quite like it. If I wanted to quickly generate an overview of a niche topic, where I had some idea of the overall structure (maybe something generated by Claude, for example), this is definitely something I’d consider using to support my own learning. I’ve published the Page, although it’s not ‘finished’ in any meaningful sense of the word, so please calibrate your expectations accordingly.