89 | Battle of the Chatbots: Perplexity
Introduction
The first four chatbots covered in this series1 are highly similar. Each presents a conversational experience: you submit a prompt, and the chatbot responds. As highlighted in the StrefaTECH AI Apps spreadsheet, there are many variations in the capabilities and user interfaces among them, but the basic engagement is the same.
Today we’ll explore a somewhat different app, Perplexity AI. It’s something of a hybrid that combines the best of traditional search engines like Google with conversational AI. Its popularity is increasing rapidly, and its approach has spawned a new competitive landscape, with apps mimicking the Perplexity approach.
What is Perplexity AI?
At its core, Perplexity AI is a search engine with a conversational twist. Instead of entering keywords and sifting through a list of results, users can engage in a back-and-forth dialogue with the AI, asking follow-up questions and receiving tailored responses in real time. The interface resembles a modern messaging app, making the experience feel more natural and user-friendly. Perplexity’s marketing positions it as an “answer engine,” an apt description for the experience as contrasted with search engines like Google.
Founded by a team of tech industry veterans in 2022, Perplexity AI was born out of a desire to bridge the gap between the wealth of information available online and our ability to access it in a meaningful way. Unlike traditional search engines that rely on keyword matching, Perplexity AI employs advanced natural language processing and machine learning algorithms to understand the intent behind each query. This allows it to provide more accurate and comprehensive responses, often anticipating and addressing related questions before they're even asked.
The Hybrid Model - Between Google Search and ChatGPT
One of the most compelling aspects of Perplexity AI is its ability to combine the strengths of traditional search engines and conversational AI assistants like ChatGPT. When you prompt Perplexity, its first step is to decide whether to do a real-time internet search (akin to a search engine) or to reference its embedded AI large language model (similar to a chatbot). I asked Perplexity to explain:
My prompt: How do you decide whether to go to the real-time internet for information, or whether to use your embedded LLM?
Perplexity’s reply: To decide whether to access real-time internet information or use an embedded Large Language Model (LLM), the choice depends on the freshness and nature of the data needed for the response. When dealing with time-sensitive queries that require the most recent data, accessing the real-time internet is crucial. This approach involves querying a search engine, retrieving information, summarizing it, and providing context to the LLM to generate accurate responses based on up-to-date evidence. On the other hand, if the data required is internal or private to an organization, using an embedded LLM with a retrieval augmented generation (RAG) technique is more suitable. RAG allows for adding external data to prompts and is ideal for scenarios where access to private data sources is necessary. Therefore, the decision between real-time internet access and using an embedded LLM depends on the timeliness and source of the information needed for a specific query.2
Perplexity AI in Action
To illustrate the power of Perplexity AI, we’ll use a real-world example. I’m leaving (today!) for a getaway to Budapest and Vienna.3 Some of the activities planned for the Budapest stint won’t be as much fun if the weather’s sour4, so I wanted to check the weather forecast before buying tickets. My experience with weather apps is that they can have highly varied forecasts, so I asked Perplexity for the forecast, which should blend information from a number of sources. My prompt sequence was:
What's the weather forecast for Budapest Hungary this week?
How about the weekend?
And next week?
The response to the first prompt is below.
A search engine (e.g., Google) could respond to the first prompt. However, because search engines don’t have a memory of the conversation taking place, they wouldn’t be able to provide appropriate responses to prompts 2-4. Each of those would be answered as a standalone question—for example, if I type “How about the weekend?” into Google after its response to the first question, I get links to various sites with things to do in Troy, Alabama (where my computer and I are at the moment!).
As a note, Google search is working to incorporate some of this conversational capability, so the difference between it and tools like Perplexity may narrow as the competition heats up.
On the other hand, an AI chatbot also could provide this information and carry on the conversation. However, the citations are either non-existent (e.g., ChatGPT) or much less prominent. For more complex research topics, the differences are even greater.
Perplexity “Focus” — Controlling Its Sources
Perplexity offers a notable feature that gives you the choice of what sources of information it will use, called its “focus.” When you begin a new chat, you can pull down the Focus menu to choose from All, Academic, Writing, Wolfram|Alpha, YouTube, and Reddit. I’ve found it to be particularly useful for searching YouTube and Reddit; having the ability to carry on a conversation can be addictive!
But Does It Hallucinate?
In short, yes. One of the most frustrating aspects of generative AI is that you can’t trust the response to be factual. Tools like Perplexity that cite sources seem to be reliable—and are indeed more reliable than chatbots—but information still may be fabricated. When Perplexity provides its response, it combs through a set of sources and summarizes the information it deems to be most relevant to the prompt. However, because its underlying engine employs standard generative AI techniques, it may include plausible but wrong information in its responses, just like other genAI tools. Similarly, if you submit the same prompt to Perplexity repeatedly, you’re likely to get different responses.
Other Players in the Field
While Perplexity AI is certainly a standout in the realm of conversational search, it's not the only player in the game. Tools like You.com and the Arc web browser (iOS only) have also emerged as contenders, each offering its own unique spin on combining search and conversation.
While these tools share similarities with Perplexity AI in their aim to enhance online search, they differ in their approaches and specific functionalities. Perplexity AI stands out with its focus on seamless conversational interaction and its ability to draw from a wide range of vetted sources, ensuring the accuracy and relevance of the information it provides.
As with every topic and tool in the AI space, though, the capabilities and players will be broadening and improving.
Conclusion
As the digital world continues to evolve, tools like Perplexity AI are paving the way for a more intuitive and engaging online experience. By bridging the gap between traditional search and conversational AI, Perplexity AI offers a glimpse into the future of information discovery, where knowledge is no longer confined to static pages but flows through dynamic, personalized interactions.
It’s a tool worth trying, though it does make it yet more complicated to…
Make Good Choices
Perplexity cited these sources in its response:
This getaway may interrupt my cadence of 2-3 StrefaTECH articles each week. Stay tuned!
Picture being outdoors at the Széchenyi Thermal Bath in a thunderstorm on a cold day. Ick!