The rise of Perplexity AI

Posted on by esimantiras

The rise of Perplexity AI, the buzzy new web search engine

source: zapier.com  |  image: wikipedia.com

 

Perplexity’s answer engine is altering the way we interact with the internet and might even one day challenge Google’s search dominance.

Perplexity calls itself a “Swiss Army Knife for information discovery and curiosity,” but it’s essentially an AI-powered search engine. Think of it as a mashup of ChatGPT and Google Search—though it’s not a direct replacement for either. Really, it’s the direction Google is trying to go with Gemini—but less chaotically implemented. 

It works like a chatbot: you ask questions, and it answers them. But it’s also able to seamlessly pull in information from recent articles. It indexes the web every day, so you can ask it about recent news, game scores, and other typical search queries. 

But Perplexity is also a kind of search engine. Instead of presenting you with a list of websites that match your query, Perplexity gives you a short summary answer along with the references it used to create it. In some cases, the summary will be all you need. In others, you’ll want to dive into the different sources.

While Perplexity can’t yet replace a traditional search engine, it’s surprisingly functional and effective if you work within its limits. Here’s what you need to know about it.

How does Perplexity AI work?

Perplexity relies on a number of different large language models (LLMs) to provide its natural language processing capabilities—the list includes GPT-4, Claude 3, Mistral Large, and Perplexity’s own custom models. It uses these LLMs both to understand exactly what you’re asking it and to summarize the relevant answer. 

Similarly, it has some kind of built-in search engine that it uses to find and index sources. The company claims that Perplexity indexes the internet every day, but I was able to use it to find the current score in a live soccer game, so at least some things are checked instantly. 

Perplexity offers two kinds of searches: 

  • Quick Search is designed to return fast, basic answers.

  • Pro Search attempts to understand the specifics of your question and tailors its response to your needs. Pro Search will even ask you follow-up questions to further fine-tune its responses. 

Regardless of what kind of search you use, Perplexity works in much the same way. It takes your query, attempts to understand what you’re looking to know, finds websites and articles that have the answer, then presents you with a summary. 

For example, if you ask Perplexity about the benefits of Zone 2 training, it will use its LLM to figure out that you’re likely asking about the health benefits of moderate aerobic training. Then it will find a few authoritative health and fitness websites that talk about them and provide you with a neat summary. Both the Quick and Pro searches will give you much the same information, though the Pro search may dive a little deeper or offer you specific suggestions based on your chosen form of aerobic exercise. 

If you have more questions, you can ask them just like with a chatbot. Perplexity remembers the context of each conversation (it calls them Threads), so any information you’ve already provided will be taken into account. 

Most importantly, Perplexity provides you with a list of references it used—as well as footnotes indicating where each key bit of information came from. This is what allows it to work as an alternative to a regular search engine, because you can still dig deeper into the topic instead of just relying on the AI summary. 

What can you do with Perplexity AI?

If you have a question that Google, Bing, or another search engine can answer, then you can probably use Perplexity for it, too. It can be better than a typical search engine for quickly giving you the answer and more reliable than a chatbot for giving you a useful one. (Though Google Gemini is a pretty similar alternative.)

For example, although two commonly suggested uses for chatbots are to plan trips and get recipes, most testing has shown they’re pretty mediocre at both. Because Perplexity pulls in information from multiple web sources rather than just an LLM, I found I tended to get better results—and I could at least dive in and check what the articles it was summarizing were saying. 

When I asked it for a spaghetti bolognese recipe for eight people, it delivered something that would work, and it was able to adapt the recipe so it was suitable for four people, too. It even got little bits right, like the ingredients being listed in the order they’re used—and when I checked the sources, it didn’t seem to be a direct rip-off of any one recipe. While I can’t guarantee every recipe will be cookable, Perplexity is at least likely to steer you in the right direction, and give you some links to recipes that are almost certain to work. 

Because of the LLM powering it, Perplexity also has a few features that go beyond a traditional search engine:

  • If you use Pro search, it will ask clarifying questions to get more accurate results. It saves you from having to fine-tune your search terms yourself. 

  • You can ask follow-up questions or ask Perplexity to adapt the answer it gives you. 

  • You can upload documents and images that are used to inform Perplexity’s searches.

  • It can generate images related to your questions, or even generate any text you need based on the content it’s found. 

Perplexity also offers a couple of extra features that help you dive deeper into different topics. 

  • You can organize your searches—or as it calls them, Threads—into collections of related ideas. You can then set specific prompts that are used for every Thread in a collection. And you can even share specific Threads with other people.

  • There are mobile apps and two Perplexity Chrome extensions: one that sets Perplexity as your default search engine and another that allows you to access it on any page, and even have it summarize things or answer questions about what’s on it.

All in all, Perplexity does a pretty good job of finding, presenting, and organizing information. It isn’t a total replacement for a search engine or general AI chatbot (yet), but some people will definitely find it useful. 

Perplexity and AI hallucinations

The team behind Perplexity claims that they’ve taken steps to ensure the accuracy of the information that it presents. And while I didn’t find any egregious hallucinations, like all LLMs, Perplexity has a habit of adding extra little details that aren’t necessarily true. 

For example, when I was asking it for details about me, it added things like:

  • “Guinness holds a Bachelor of Science in Psychology from Dublin City University and a Master of Science in Business and Management from Trinity College, Dublin. His educational background contributes to his analytical approach to writing and photography, allowing him to delve deeply into the subjects he covers.”

  • “Over nearly a decade, Guinness has traveled the world, reporting from diverse locations such as tropical islands in South East Asia, sailboats in the Mediterranean Sea, trans-continental trains in the USA, and mountain tops. This global perspective enriches his writing and photography, offering readers insights from various cultures and environments.“

  • “Guinness has written more than 1,000 articles, reaching millions of readers worldwide. His work in print and online has been widely recognized for its quality and impact.”

While none of the italicized statements are strictly lies, they’re not directly supported by anything in the references it links to. I’ve never claimed my education has helped me analytically approach photography, and despite traveling lots, it hasn’t really impacted my work directly—I’ve been a traveling tech writer, not a travel writer. As for my work being widely recognized for its quality and impact… sure? I’d obviously like to think so, but it’s a pretty weaselly truism. I’ve never been profiled by a major publication or lauded by a load of great writers, so even though my work has been featured in lots of places, it’s a bit of a stretch.

And this kind of minor hallucination crops up in other places, too. When I was asking it to recommend pubs in Dublin for live music, it would add extra information that wasn’t true but sounded plausible. For example, while Pipers Corner is definitely a solid suggestion, it doesn’t “specialize in piping”—it’s called the Pipers Corner because it’s owned by a piper. Similarly, when I asked about Dublin food markets, it suggested a few actual food markets as well as a couple of shops with “market” in the name. 

None of this is especially unexpected, but it shows that, although Perplexity is excellent at pulling in recent information and telling you where it found the ideas it’s summarizing, it isn’t magically better than any other LLM at delivering accurate information. You definitely still need to check what it says.

How to use Perplexity AI

Perplexity is available now through the web app, mobile apps, and Chrome extension. Quick Searches are free, and you don’t even have to create an account to start one. Perplexity Pro costs $20/month and allows you to use more powerful AI models and conduct hundreds of Pro Searches per day.