Click the Hard AI to follow us
Hard·AI
Author | Hard AI
Editor | Hard AI
To be frank, the impact of habits on a person is immense.
Although I have been using AI for a long time to help me with many tasks, there is one scenario where I am still very path-dependent, or rather the power of habit has prevented me from changing.
That is: searching.
In fact, since the internal testing of NewBing began last year, I have been using it extensively, but later experiences and accuracy really made me want to complain.
Eventually, I gradually returned to Google and Baidu.
Until recently, a piece of news led me to discover a treasure AI, completely defecting from traditional search to the AI search camp—
Old Huang said that the search engine he uses every day is not Google, butPerplexity.
So amazing? What is this company all about? Let’s check—
Perplexity is an AI question-answering engine.
Similar to NewBing? Is it really that powerful… Let’s check again, and I was shocked—
Wow, this guy not only saw its valuation increase nearly10 times in a year (soaring to 1 billion USD),
but also quietly ranked seventh in the top 50 generative AI monthly list released by A16Z, surpassing major AI applications like Claude, Midjourney, and 11-Labs.
So, why is this phenomenal AI-native application so powerful? How does it differ from traditional search engines like Google?
01
Question-Answer Engine: Disrupting Search in the AI Era?
We can start with a question to see the main differences between it and traditional search engines.
For example, to provide my friends with a better video experience, I decided to upgrade my microphone.
Let’s first search on Google for a microphone under 100 yuan that sounds like it costs 1000 yuan. Well, Google gives you a bunch of results, and just reading through this page would take at least half an hour.
However, if we ask Perplexity the same question, it will subtly refuse you: No, don’t think about it, you won’t find it.
From this simple comparison, we can see that:
Traditional search engines, like Google, return a large number of webpage links. But are those webpages really what we want? Clicking one might be an advertisement, another might be clickbait. You may need to open many links, read carefully, and discern before you can find the answer.
In contrast, with the support of large models, Perplexity can help us search, read, and summarize information from different sources, quickly distinguish the most valuable information, and organize it logically to give you a complete, summary, one-stop answer—
This is something traditional search engine technology cannot achieve.
This is also why search is a definite track that will inevitably be thoroughly transformed by large model technology.
02
Simplified and Customized for Search
Next, let’s take a look at its key features.
Opening the Perplexity page, the design is very simple, just a search box, typical of a search engine style, and its key features are concentrated in this small box.
First, the “Focus” option in the lower left corner is easy to understand; it allows the large model to focus on a specific type of information source, such as: academic papers, Reddit discussions, or YouTube videos. If you choose Writing, it won’t connect to the internet, essentially having a direct conversation with the large model.
The “Attach” option next to it is its file upload feature. This function is now standard for AI tools.
The “Pro” option in the lower right corner is a more advanced AI search. It uses a step-by-step approach, asking you further questions to help the AI better understand what you really want, thus being more accurate.
After the search is complete, the Perplexity search results include the following parts:
“Sources”: lists the sources cited for the answers, supporting clickable links; in the example below, Perplexity cited a total of 5 different sources.
“Answer”: a structured answer derived from integrating content, containing references; for example, this explains the positioning, pricing, parameters, and other key information about the Xiaomi SU7.
“Related”: after completing the answer, Perplexity also provides several related questions you might be interested in, which you can directly click to ask further questions or continue submitting text.
In addition to search, Perplexity also offers a “Discovery” section, which acts as a news aggregator where users can see the latest trending news.
It is worth mentioning that Perplexity also allows you to freely invoke GPT-4 and Claude 3 as built-in engines—of course, you can’t get it for free; it costs 20 bucks a month.
03
Keep Going! Young Man
As for Perplexity, its drawbacks are also obvious.
First, since it is a generative AI, hallucinations are inevitable. Although Perplexity tries to ensure the relevance and accuracy of results by searching multiple information sources, it still requires careful fact-checking.
To handle this, Perplexity needs strong RAG technology.
Secondly, although Perplexity leads traditional search engines in search experience, its functions and applicable scenarios are still limited. For example, it cannot cover all information on the internet like Google and lacks practical functions like map navigation.
Therefore, it is still too early to say that Perplexity can overthrow traditional search engines like Google. To become the new standard for search engines in the AI era, Perplexity must continuously improve in AI capabilities, information coverage, and other aspects.
Hard·AI
