Artificial intelligence startup Perplexity AI has raised tens of millions of dollars from Jeff Bezos and other notable tech investors with a mission to take on Google in the information-search business.
But the company's AI-driven search chatbots have already faced challenges, with some news outlets raising objections to its business practices. The company also faces competition from tech giants Google and now Apple, which are increasingly building similar AI capabilities into their core products.
Perplexity CEO Aravind Srinivas is defending the company after it published a summary article that contained similar information and language to a news article. Forbes Forbes published the investigative piece but did not cite the media sources or ask for permission. Forbes later said it had found similar “copycat” articles that had been plagiarized from other publications.
The Associated Press also found that another Perplexity product feature fabricated fake quotes from real people, including one that quoted a former town official on Martha's Vineyard saying he didn't want the Massachusetts island to become a marijuana destination.
“I never said that,” said Bill Rossi, a former elections board member for the island city of Chilmark.
Srinivas told The Associated Press that his company strives to build good relationships with news publishers so that their news content “reaches a wider audience.”
“We can definitely coexist and help each other,” he said.
Asked about Forbes, he said his company's products “have never plagiarized anyone's content. Our engines aren't trained on other people's content,” in part because the company simply aggregates what other people's AI systems generate.
“We are actually an aggregator of information, providing it to people with proper attribution,” Srinivas said, but added: “Forbes correctly indicated that they would like us to highlight sources more prominently. We took that feedback immediately, fixed it that same day, and now sources are highlighted more prominently.”
Perplexity also said this week that it is exploring revenue-sharing partnerships that would pay news publishers a portion of Perplexity's advertising revenue each time their news content is referenced as a source.
Randall Lane, chief content officer at Forbes Media, called the controversy a “tipping point” in the discussion about AI.
“This is a case study of where we're headed,” Lane told The Associated Press. “If the people leading this effort don't have a fundamental respect for the hard work of doing original journalism and keeping people informed with value-added content, then we're in big trouble.”
Lane, who calls himself an “AI advocate” because he believes AI technology can help many news organizations become more efficient, said the dispute between Perplexity and Forbes is important because it's “a metaphor for what happens when the people who control AI don't respect the people who do its work.”
Lane said that while Perplexity describes itself as a search engine, it “acts like a media company, publishing stories that are exclusively reported by Forbes.”
“It was very dishonest across the board, and what we heard wasn't, 'Oh, oh, this failed, we've got to do better,'” he said. “Instead, they were putting out more content, tweaking their model a little bit, treating journalism like it was just a manufactured commodity.”
Srinivas, a computer scientist and former AI researcher at OpenAI and Google, co-founded Perplexity in the summer of 2022, shortly before AI image generation tools Stable Diffusion and OpenAI's ChatGPT began to spark public interest in the potential of generative AI.
Inspired in part by his childhood love of Wikipedia, he described Perplexity to The Associated Press as “kind of like a mix between Wikipedia and ChatGPT,” saying it can instantly answer people's questions without the “messy, confusing” results that Google's traditional search results have.
“You ask a question, you get an answer from a clear source, and you end up with three or four suggested[follow-up]questions,” he said of Perplexity. “That way people's minds are freed from distractions and can be focused on learning and deeper exploration.”
The company sells subscriptions for premium features and plans to launch ad-based services as its user base grows.
“We're not currently profitable as a company, but we're running a more sustainable business than basic model companies because we're not training our own basic models, which requires massive computing power,” he said.
Perplexity relies on existing AI large-scale language models built by OpenAI, Anthropic and Facebook's parent company Meta Platforms, and “post-trains” them.
“We're training them to be really good summarisers,” he said.
It's not always clear where the summarized information comes from. Perplexity's feature called “Writing” (which allows users to “generate text or chat without searching the Web”) generates lengthy, unsourced commentaries, often in the style of news articles. When an Associated Press reporter tested the feature by asking him to write about a marijuana shortage on Martha's Vineyard, it produced a 465-word document that resembled a news article and included fabricated quotes from a former town official and another real person.
The AP did not repeat the misquote to avoid spreading misinformation. Srinivas said Perplexity's writing feature is a “minor use case” aimed at helping people write essays or correct grammar when primary sources aren't needed. He said the feature is “prone to hallucinations” — a common problem with AI's large-scale language models — because it's not tied to the web search capabilities of Perplexity's core product.
“There's no question that generative AI is transforming journalism, content creation and search,” said Sarah Kreps, director of Cornell University's Technology Policy Institute.
She pointed to approaches like Google's new Perplexity, which summarizes answers based on information it gleans from crawling the web, which also led to misinformation and forced Google to tweak the product after it was released to the public.
“But their entire advertising model is based on driving people to a website,” she said in an email. “Why would people go to a website when the output of their AI can give them a one-stop answer?”
“A lot of people have been referred to us by Perplexity, and it's good to see that we're getting referrals from new players on the internet,” Srinivas told the Associated Press.
For now, many of the benefits may be aspirational: Perplexity's global user base has grown rapidly this year, reaching more than 85 million web visits in May, according to data from Similarweb, but that number is tiny compared with the billions of users of ChatGPT and other popular platforms from Microsoft and Google.
Steven Lind, an associate professor at the University of Southern California's Marshall School of Business, said the controversy speaks to “uncertain and challenging times” for online content creators in general, and journalism in particular, because aggregators can't function without publications like Forbes.
Using AI as an integration tool will help disseminate information widely “until the original is gone,” he said.
“Entire enterprises and applications are doing this same thing. They're rolling out applications for industries that maybe aren't their expertise, so they roll out new services without fully considering the impacts, best practices and safeguards,” he said.
Lind said it's good to see companies like Perplexity “at least taking some steps to course correct when there's backlash from the industry and from users,” but he added that some changes should have been built in from the start.
____
AP
