Original video in Chinese.
Key Takeaway
- Perplexity is currently the best Q&A engine, and its answer quality and user experience are better than products like ChatGPT.
- As the starting point for knowledge, Perplexity emphasizes building trust (clear source citations), helping users ask questions (expanding search keywords, guiding prompts), and providing one-stop service (Pages feature).
- The Pages feature combines search with content ideation and can help users generate article outlines and supporting information.
- Perplexity achieved its leading position in Q&A engines by fine-tuning GPT models, using other large language models, and developing its own open-source model, combined with powerful RAG technology.
- Perplexity also offers the Focus feature (specific search directions) and the Space feature (knowledge base), further expanding its use cases.
Full Content
ChatGPT recently launched a search feature, but I’m still going to keep using Perplexity. Because as a Q&A engine, or the next-generation search engine, Perplexity still has the best answer quality and user experience. Building a good Q&A engine requires a lot of domain knowledge and also involves engineering challenges; it’s not something you can handle just because you have a strong model.
As for OpenAI, honestly, I’m not very optimistic about them. I even complained about this in my community before, saying that OpenAI now feels like Huang Lei: jack of all trades, master of none. They want to do too many things:
They want to develop foundation models, build applications, and also create an app store and ecosystem. A series of features they’ve launched have all been half-measures, never really pushed through thoroughly. If you actually want to use them seriously in production, they’re still a bit short.
So over the past year, I’ve firmly chosen the combination of Perplexity and Claude. They have genuinely helped me and helped me make money. This video is basically a tutorial on Perplexity. If you haven’t used it yet, or haven’t subscribed yet, definitely keep watching.
Hello everyone, welcome to my channel. Modestly speaking, I’m one of the few bloggers in China who can clearly explain the Why and How of AI. Remember to follow me. As long as you watch even one video all the way through, you’ll have made a huge gain. If you want to connect with me, come to the newtype community. More than 500 friends have already paid to join!
Back to today’s topic: Perplexity.
There are two very certain tracks in AI applications today: one is search, and the other is knowledge bases.
I don’t need to say much about search. After large language models appeared, both general search and vertical search saw a flood of new products, and there’s no doubt a unicorn will emerge. Knowledge bases are also extremely hot. RAG as a Service, building a Knowledge Assistant, this process has been accelerating all along.
Have you noticed? Whether it’s search or knowledge bases, they’re both about knowledge, both about the discovery and flow of knowledge. That’s why Perplexity once put a line on its interface: Where knowledge begins. Too bad it’s now changed to “Unleash your curiosity,” which instantly feels much weaker.
Once you start using Perplexity, you’ll feel this: a Q&A engine is not just a large language model plus search. It’s really not that simple.
First, as the starting point of knowledge, it needs to build trust, so its behavior and results must be rigorous and proper, like writing a paper:
All sources are clearly cited, and you can trace them back very easily; the generated results are also logically clear, concise, and easy to understand.
Second, as the starting point of knowledge, it needs to help users ask questions. Perplexity’s founder said something in an interview that left a deep impression on me:
Their biggest enemy is not giants like Google, but the frustrating fact that users don’t know how to ask questions.
Most users haven’t figured it out at all; even if they have, expressing it accurately is another barrier.
So Perplexity has strengthened its product guidance. For example, after the user finishes entering a question, if it isn’t specific enough, they’ll give a few options to probe the user’s real intent. Another example: not long ago, they also added an autocomplete feature.
Third, as the starting point of knowledge, it also needs to take care of the journey that follows—it shouldn’t just be a starting point; it should, as much as possible, become a one-stop presence.
The follow-up suggested questions feature is simple, but very practical. After all, with so many questions, there’s always going to be one that hits the mark.
The Page feature launched some time ago is very clever. Starting from one question, you keep asking, keep expanding the logic, and eventually form something like an article outline plus supporting information.
If the need isn’t complicated—for example, if you just want a travel guide or something like that—then this finished product is basically ready to use as is. If you’re writing an article, then this basically completes the initial brainstorming—the overall logic of the article is there, and the supporting material is there too.
I think the Page feature has taken the process we’re used to—searching while thinking—and made it concrete and productized. If they keep refining it, maybe one day it really will achieve the effect of directly producing a finished draft.
So, once you understand these features of Perplexity and then compare them with large language model products that include search, you’ll find they really aren’t the same thing; they feel like two different species. I strongly believe that adding AI on top of search and adding search on top of AI are two different product forms. There was even a period when, after unsubscribing from ChatGPT, I used Perplexity as a replacement. Because it also has the ability to generate text directly.
Many people may not have paid much attention to the Focus button. Once you click it, you can set a specific search direction. For example, you can search specifically YouTube or academic papers, and so on. There’s also a Writing option, which doesn’t search at all, but instead answers the user’s question directly based on the model’s existing knowledge. If you want to change the model, go into settings and choose one. In addition to the GPT series, you can also choose the Claude series. All the most advanced models are available here.
Through the internet, most of the knowledge we can search for is public, general knowledge. There’s also a lot of domain knowledge that isn’t online. To cover this situation, Perplexity launched the Space feature some time ago, which is the knowledge base. Users can upload documents and let the model answer based on those documents.
A knowledge base is a piece of cake for Perplexity. Because a Q&A engine inherently needs strong RAG capabilities. Following this direction, I especially hope they’ll add a document management system, including hierarchical folders and tags. Let users gradually store private documents in it and complete the migration of data assets—that’s something an entry-level product absolutely has to do.
Perplexity also has some less core features, such as the Discover page. You can think of it as the news recommendation pages that Google and Baidu both have. At the moment, it doesn’t seem very interesting; let’s see whether there’s any innovation there in the future.
OK, that’s it for this episode. After watching, everyone remember to go try this currently best Q&A engine. See you next time!