I sometimes use the generative AI tool Perplexity when I’m working on We Love Translations. What do I use it for, you ask?
Perplexity for automation?
Perhaps unsurprisingly, Perplexity will not fetch book cover images based on a list of ISBNs because it was taught not to do this kind of web scraping on behalf of users. (Fair enough.)
Perplexity failed to generate a correct Excel formula for calculating 10-digit ISBNs from 13-digit ISBNs. After a couple of tries, I was still getting wrong results, so I just Googled and found a message board where someone posted an ISBN-13 to ISBN-10 Excel formula that actually works, so I’m using that. (However, I got Perplexity to explain it to me!)
For the Excel sheet where I track donations to and affiliate earnings from We Love Translations, Perplexity wrote me a giant Excel “if” statement that outputs how much my Amazon bonus for the month is already, and how much it could be, based on the day of the month and the current sales revenue so far. (I use Excel a lot, but usually not for numbers, oddly enough.)
Perplexity will NOT make a collage of images you give it based on instructions about the arrangement and background. I tried this when I was working on my We Love Translations post for Around the World in Eighty Days. Rather than returning an image with two rows containing the ten book covers I supplied, Perplexity returned an image with three rows of three “book covers” that it generated itself. Some of the covers resembled the book covers I supplied, but all of those were at least a little… off. Other covers were not based on covers I supplied. Overall, the result was hilariously bad.
On the other hand, Perplexity wrote a PhotoShop script that will take a folder of images and create exactly the kind of collage I want! No more resizing, importing, and arranging images on a grid by hand. Just choose a folder and boom! The collage is created all in one go. Amazing. (Why didn’t I do this ages ago???) My husband improved the script manually, because Perplexity somehow wasn’t able to tell Photoshop to put the images in the same order that they appear in Windows Explorer. Still, Perplexity is a big productivity boost here.
I’ve also used Perplexity to automate the task of posting memes to Tumblr.
Perplexity for debugging?
Absolutely. I had a problem with my PHP template code in my translations website. For a long time, I was able able to fill in a bunch of custom fields in a table automatically by uploading a spreadsheet, but after a while, the upload step started populating the table with two copies of the data instead of one. Then I’d have to go and manually delete half the rows. Very tedious. Couldn’t figure out what was going on. Asked Perplexity, and immediately got back an explanation about how to adjust the arguments of the relevant WordPress function call. Instant fix. So grateful!!!
Perplexity as a search engine?
Like Google, Perplexity can find answers to factual questions, but prudence dictates that such questions be limited to questions that have widely accepted answers, or answers that can be derived from some sort of survey or sampling process. I’ve had good results with answers to some of the English language usage questions I’ve encountered in my day job, such as: Is it better to write “F1-Macro score” or “macro F1 score”? (The latter.) Or: What is the difference in usage between “demands” and “demand”? (The field of economics does actually use the former, though it sounds strangely aggressive to me.) On the other hand, I’ve had mixed results asking for software suggestions… when I was looking for free desktop software for doing optical character recognition, it pointed me to open-source tools that haven’t been updated in years, and in one case I’m pretty sure a link supplied by Perplexity landed me on some sort of scam site that had taken over a domain using the name of a formerly popular tool. (Thanks a lot.)
I don’t trust Perplexity on niche topics, and publication details relating to translations of foreign books is one such topic. (If this weren’t a difficult niche, someone else would already have built something like We Love Translations before I did.) The advantage of using Perplexity is that it can read and summarize faster than I can; but it can’t find things that are hard to find, and it summarizes by mixing unrelated pieces of information together if it can’t find much that relates to the question, without any warning that that’s what it’s doing. I’ve found that it pays to visit at least some of the links supplied in the answers, just as I would if scrolling through the Google search results page. I still trust actual webpages more than Perplexity’s summaries of webpages.
Not unlike a human, Perplexity can’t prove a negative. In other words, don’t bother asking if it has found all the English versions of the title of Kafka’s Metamorphosis, because it won’t know if it hasn’t (duh!). In fact, when I told it I thought it was “missing one”, it produced over 150 pages of anxious dithering on the subject, and never changed its answer! The title it couldn’t find was Peter Wortsman’s Transformed, which I had discovered in the process of Googling. So… I’m a long way from ditching Google.
In fact I still really like Google Search, and I’m trying not to rely on the generative AI summaries Google now puts at the top of seemingly every search results page. Those summaries, for many people, serve as a substitute for reading webpages in the search results listing. Traffic to We Love Translations from Google Search is waaaaaay down, and I believe that’s one of the reasons, another being that maybe Google doesn’t trust webpages as much now that anyone can generate content quickly, therefore it’s pushing forums like Reddit, Quora, and Goodreads to the top of the search results, which, sadly, pushes my website down. (C’est la vie.)
Perplexity seems capable of useful “advanced search” behavior. Years ago, I saw a book on puns in a second-hand bookstore that had on the cover a photo of a woman’s violin-shaped back, complete with a pair of S-shaped “holes.” I didn’t buy the book. I intended to buy a cheaper copy at some point later, but I forgot the author’s name! Searching Amazon, Google, Abebooks, etc. for a book titled “Puns” naturally produces so many results that the needle gets lost in the haystack. How would I ever be able to find the book again? But then I recently ran into a list of famous photos including the one of the violin/woman. I told Perplexity about the photo and the unidentified book, and—well, you know how this story ends! The author’s name is W.D. Redfern. I remember a similar experience in 1999 or 2000, when, having failed to identify a kids’ book I remembered using a library card catalog (!), I was able to find it using an online search engine. Paradigm shift.
Perplexity for writing?
Nope nope nope nope. I would not copy-and-paste Perplexity’s output into a blog post or webpage, even as the starting point for a draft, because I strongly believe copy-paste-edit is not writing, it’s copying. (Paul Graham says writing is thinking; I agree… and why would I let a tool do my thinking for me?!) Anyway, what Perplexity is doing is essentially already copy-paste-edit: it’s taking content scraped from the web, putting it in a blender, and spitting it back out again, smoothed out a bit. Well, I know how to make smoothies, and what typically happens, at least with my blender, is that the blended drink still contains big chunks of ice. Similarly, generative AI tools may plagiarize by spitting out recognizable bits of the input data. Plus, it seems that AI tools, like blenders, don’t much care what goes in them. They’ll happily mix together things that don’t go together at all; and nobody really wants to swallow that, do they? Garbage in, garbage out!
Perplexity for translation?
Nah… I still use Google Translate. Meanwhile, however, people are publishing English-language AI translations of public domain works written in other languages. My knee-jerk reaction is that that’s a deceptive, cashgrabby thing to do, and I really wish they wouldn’t. (In contrast, I’m in favor of automated low-cost translation from the world’s major languages into languages for which translation wasn’t economically viable.)
On the other hand, I have used Perplexity to understand and improve word choice. A Chinese author wrote something about the “rationality” of a certain method, which made it sound like the method was a conscious being—clearly not the author’s intent. I asked Perplexity what Chinese word he was thinking of, and how it ought to be re-translated into English. There isn’t a great paraphrase (I think I suggested “soundness”), but the idea is that the method is “reasonable” in the sense that it was designed by someone in a rational, sensible, reasonable way.
Perplexity for editing?
I’m delighted that Perplexity can reformat Latex tables. This is a real time saver! I can tell it to add a column or remove all the boldface commands rather than squint at the code and manually move the cursor around to insert and remove little bits of text, invariably making mistakes along the way. Perplexity will do a better job, and faster. That’s not language editing, though, it’s text processing or something.
I know that people use chatbots to clean up their writing; but not every change is an improvement. I’ve never used Perplexity to rewrite a whole paragraph, to say nothing of a whole document. I’d still rather do that myself.
Perplexity for image generation?
I have used AI-generated images in PowerPoint presentations. Specifically, I created colorful portraits of Team Tantaloid, a group of pretend researchers, for the purpose of attributing English-language writing mistakes to them in a PowerPoint presentation. I also generated an image of their tech project, a kind of glowing cube thing. I didn’t use Perplexity, though; I made the cube image with Canva and the portraits with Image Creator from Designer using Copilot, which was powered by DALL-E3.
In general, AI-generated images have the same two problems that AI-generated text has: possible plagiarism or copyright violation and lack of accuracy. The quality of the output has gotten better over time, and maybe also respect for others’ intellectual property has too, if only because of lawsuits like Getty vs. the creators of Stable Diffusion. Still, I think we have a ways to go.
…Perplexity as a verb?
I mean, we never thought Google would be a verb, did we, but here we are. (As Calvin says to Hobbes, “Verbing weirds language.”) However, the name of the world’s most popular search engine is only two syllables. Nobody in their right mind would bother to enunciate, “I’m going to Perplexity this,” which is not only long but also sounds ungrammatical, probably even to people who don’t care about grammar like I do. Maybe it works as “Perplex this”, but “perplex” is already a verb—one with two annoying consonant clusters that make it hard to say—and it doesn’t mean “clarify”; quite the opposite. Meanwhile, “Perp this” is clearly unacceptable… So really the best I can do is “ask Perplexity”, which clocks in at five syllables. Oh well. It’s not worse than “ask ChatGPT”.
