Perplexity AI is Twitter Minus the Drama

In a recent podcast with Lex Fridman, Perplexity AI chief Aravind Srinivas compared his company with social media platforms such as X (formerly Twitter) and Threads, insinuating that these platforms provide news alongside drama.

“Twitter is great; it serves many things like human drama in it, there’s news, there’s knowledge you gain, but some people just want the knowledge, and some just want the news without any drama,” he said.

Srinivas said that the solution may not even be starting another social app. “Meta started Threads and tried to say, ‘I want to start Twitter without all the drama,’ but that’s not the answer,” he said, adding that the answer is to try as much as possible to cater to human curiosity but not to human drama.

Interestingly, when Perplexity started, it scraped Twitter data. However, when Elon Musk took over, he restricted free Twitter API access.

“We knew that relying on Twitter search was not scalable or feasible for us because Elon was taking over and he was very particular about shutting down API access. So, it made sense for us to focus more on regular web search,” said Srinivas.

Notably, Musk later introduced his own chatbot, Grok, inside X

Srinivas further said that Perplexity AI is trying to change the way people discover knowledge, adding that as of now people mostly use Google Search to gather information.

“It’s evident that around 30 to 40% of Google Search traffic seeks direct answers, while the rest explores deeper inquiries, much like what we’re providing.”

He said that Perplexity AI users are able to ask deeper questions, which is not possible on Google Search. “I just believe we’re working towards neither a search engine nor an answer engine but ‘knowledge discovery’. That’s the bigger mission, and that can be catered to through chatbots, answer bots, or voice factor usage.”

Perplexity AI recently launched a new feature called Pages, where users can ask questions and create, organise, and share information. Simply put, they can search any topic and instantly receive a well-structured, and beautifully-formatted article.

Moreover, they can publish their work and share it directly with their audience with a single click.

“Through the Pages project, we’re working on enabling people to create new articles with minimal human effort. I hope the insight from your browsing session and the query you asked on Perplexity will be useful to others as well. One should be able to broadcast what they learned from a single Q&A session on Perplexity to the rest of the world,” he quipped.

How Perplexity AI works

Perplexity AI utilises the RAG technique to extract data from the internet. “The principle at Perplexity is you’re not supposed to say anything that you don’t retrieve,” said Srinivas. He explained that their model is designed to avoid generating anything that isn’t present in the document, ensuring factual grounding.

The company has also built a crawler to index websites. “We have the Perplexity bot, Bing bot, GPT bot—a bunch of bots that crawl the web,” said Srinivas. Lately, there have been a few instances where Perplexity AI has been reportedly accused of directly ripping off content from websites.

He explained that they have to use headless rendering because websites are more modern these days. “It’s not just HTML, there’s a lot of JavaScript rendering, and you have to decide what you really want from a page,” he said.

He added that websites often have robots.txt files, and one should respect the delay time so that they don’t overload their servers by continuously crawling them.

Prompt Engineering is Dead

Srinivas hinted that the company’s upcoming products might remove the necessity for users to frame questions accurately and precisely.

“I believe the whole prompt engineering thing, like trying to be a good prompt engineer, is not going to be a long-term thing. I think you want to make products work where the user doesn’t even ask for something, but you know that they want it, and you give it to them without them even asking,” he said.

“I don’t even need them (users) to type in the query. They can just type in a bunch of words and that should be okay. That’s the extent to which you got to design the product because people are lazy, and a better product should be one that allows you to be more lazy, not less,” he explained.

Srinivas added that the biggest enemy to them is not Google. Instead, it is the fact that people are not naturally good at asking questions.

“Every person in the world is curious, but not all of them are blessed to translate that curiosity into a well-articulated question. There’s a lot of human thought that goes into refining your curiosity into a question, and then there’s a lot of skill in making sure the question is well prompted for these AIs,” he said.

Targets Japan

SoftBank Corp recently announced a partnership with generative AI search startup, Perplexity. Beginning June 19, 2024, SoftBank offered customers of its SoftBank, Y!mobile, and LINEMO mobile services a one-year free subscription to the premium version of Perplexity’s AI answer engine.

Srinivas recently visited Japan, where he met developers and users. Despite facing accusations of plagiarism and copyright infringement, as well as increasing competition from Google and OpenAI, Perplexity AI is determined to stay relevant and continue its growth.