AI Search: A New "Three Kingdoms Battle" Begins. Who Will Be the Winner in This Race?
The Somewhat Satisfactory SearchGPT
Shortly after SearchGPT launched, Silicon Rabbit obtained a test account. After using SearchGPT for two consecutive weeks, the biggest impression was: ### The search speed is really fast!
Compared to the search speed of Perplexity's free account, Bing Search, and interaction with ChatGPT, SearchGPT's search speed is lightning fast, comparable to Google's keyword search.
Apart from this, Silicon Rabbit believes that SearchGPT's current limitations are:
-
The user interface is relatively simple, somewhat crude compared to Perplexity;
-
The sources of retrieved and synthesized information seem limited; for open-ended questions, it doesn't seem to provide as much valuable information as Google.
After two weeks of use, Silicon Rabbit believes that SearchGPT and Google Search can be used in combination. For example, for questions that need quick answers, ask SearchGPT for rapid responses. For questions that require brainstorming, comprehensive information from multiple aspects, exploration, and in-depth understanding, one still needs to turn to Google. ### So at this stage, it's still impossible to completely break away from reliance on Google.
SearchGPT is good at handling questions like "Perplexity's funding and valuation information" - questions with clear direction and limited answers.
The left menu bar shows the three functions currently supported by SearchGPT: search, view indexed links, and search images. For this question, SearchGPT provided 20 search results, which is somewhat different in scale compared to Google's search volume.
Additionally, SearchGPT has some of ChatGPT's capabilities. For instance, it can search for the latest status of a series of input company names, such as IPO/M&A/private equity status.
This is an ability that Google doesn't possess.
However, for more open-ended questions, or when you want to see more related information and explore further, SearchGPT's retrieved information resources are relatively limited (generally providing 10 indexed pieces of information, while Google's first two pages of search results provide twice the amount of information), making it difficult to meet this need.
But SearchGPT has prepared a trump card to leverage users - if you choose to set SearchGPT as your search engine, the link will redirect to Google Chrome's mini-program app store (Chrome Web Store). After choosing to integrate it into the browser, Google Chrome's originally default Google Search will change to SearchGPT.
To some extent, SearchGPT directly "altered" Google Chrome's default Google Search. ### Undoubtedly, this move enhances the convenience of using SearchGPT, "snatching food from the tiger's mouth" of Google.
Perplexity's Strong Entry
Perplexity offers more "certainty" and "richness" in search experience and is viewed by the outside world as a challenger to Google.
Founded in 2022, Perplexity was co-founded by former OpenAI researcher Aravind Srinivas, former Meta scientist Denis Yarats, and other partners.
Whether SearchGPT has drawn inspiration from Perplexity is unknown, but in terms of specific functional experience, the two are very similar. For example, Perplexity uses AI to provide users with more direct search answers rather than simply providing website links.
Moreover, all of Perplexity's answers are followed by information sources, much like scholars citing references in their papers. For instance, if a user inputs "Paris Olympics," the displayed answer will be followed by links to reports from renowned media outlets such as Reuters and Bloomberg.
On Perplexity, you can also see a series of related questions, such as "The medal-making process for the Paris Olympics" or "The competition schedule for the Paris Olympics." These are called "follow-up questions," which users can read further if interested.
Perplexity also has a function to enhance "certainty" by allowing users to limit their search scope to specific databases, such as requesting it to restrict the search range to YouTube, Reddit, or specific academic journals.
In 2023, Perplexity's monthly active users exceeded 10 million, with over 5 billion queries provided. The goal for 2024 is to increase both figures tenfold.
Aravind Srinivas once claimed, "Google will be seen as traditional and outdated, while Perplexity will be the next-generation and future product."
Google's response to this impact can be described as quite poor.
The Common Problem of the Three Players: Information Accuracy
At this year's annual I/O Developers Conference in May, Google prominently announced the introduction of AI features into its core search business, launching a function called "AI Overviews."
In simple terms, when users perform a search, AI compiles a summary that is displayed at the top of the web page for priority viewing. Users don't need to click on multiple links to quickly get answers. At the time, Google CEO Sundar Pichai called it the biggest change in the search field in 25 years.
However, the ideal is plump, but reality is bony. Google soon terminated AI Overviews. The reason was that it displayed ridiculous and incorrect answers when responding to some users' questions, leading to fierce criticism from users.
The most discussed issue was that when answering the question "cheese doesn't stick to pizza crust," AI Overviews suggested "adding an eighth cup of non-toxic glue to the toppings."
In fact, SearchGPT is also facing issues with information accuracy.
A user input "August music festival in Boone, North Carolina" into SearchGPT, and the first result showed the Appalachian Summer Music Festival, stating that it would host a series of art events from July 29 to August 16 this year.
However, after verification, the user found that the festival actually starts on June 29, with the last concert held on July 27. In other words, there would be no events from July 29 to August 16, clearly indicating that SearchGPT provided an incorrect answer.
Not just SearchGPT, Perplexity also faces issues with incorrect information.
Previously, Wired's investigation found that Perplexity used third-party content crawlers to scrape content from websites that do not allow crawling. Additionally, Wired pointed out that Perplexity's answers would incorrectly summarize article content and even spread misinformation.
Aravind Srinivas's response was very firm, directly retorting that "the problem lies with the third-party crawlers, not Perplexity itself."
However, when publishers publicly criticized Perplexity, it softened its stance. Previously, Forbes claimed that Perplexity stole text and images, infringing on their copyright. Aravind Srinivas responded by saying that the product features had flaws and were being improved through more feedback.
Aravind Srinivas also stated that the company is working on reaching revenue-sharing agreements with publishers to address the emerging issues.