Another problem with a system like ChatGPT is that its responses are only based on the data it was trained on. Retraining a model may well cost millions of dollars due to its size and scale of data. You Chat is confused when asked about the latest sports scores, but he knows what the weather is like in New York at the moment. Soher doesn’t want to reveal how up-to-date information is included, seeing it as a competitive advantage.
“I think a lot of these chat interfaces right now are a lot better than the search experience in some ways, but a lot worse in others,” says Socher. “We are working to reduce all of these problems.”
Aravind Srinivas, founder and CEO of search startup Perplexity AI, who previously worked at OpenAI, says the challenge of updating a system like ChatGPT with fresh information means they have to be combined with something else. “Alone, they can never be good search engines,” he says.
Saam Motamedi, a venture capitalist at Greylock Partners who invested in AI-powered search company Neeva, says it’s also unclear how compatible chat interfaces are with search engines’ main revenue model — advertising. Google and Bing use search queries to select the ads that appear at the top of the list of links provided in response. Motamedi suspects that new forms of advertising may need to emerge for chat-style search interfaces to be viable, but it’s not entirely clear what that will be. Neeva charges a subscription fee for unlimited search without ads.
The cost of running a model like ChatGPT at Google’s scale could also prove problematic. Luis Chezé, co-founder and CEO of OctoML, a company that helps companies reduce the cost of deploying machine learning algorithms, estimates that performing a ChatGPT search can be 10 times more expensive than a Google search because each answer requires running a large and complex artificial intelligence model .
The scale of the ChatGPT craze has surprised some programmers and artificial intelligence researchers familiar with the underlying technology. The algorithm behind the bot, called GPT, was first developed by OpenAI in 2018, and a more powerful version, GPT-2, was introduced in 2019. It’s a machine learning model designed to capture text and predict what’s next, which OpenAI has shown can perform impressively when trained on large volumes of text. The first commercial version of the technology, GPT-3, has been available to developers since June 2020 and can achieve many of the things that ChatGPT has been praised for recently.
ChatGPT uses an improved version of the base algorithm, but the biggest leap in its capabilities comes from OpenAI forcing people to provide feedback to the system about what makes a satisfactory response. But like previous text generation systems, ChatGPT is still prone to reproducing biases in its training data, as well as “hallucinating” plausible but incorrect results.
Gary Marcus, professor emeritus at New York University and critic of the AI ​​hype, believes that ChatGPT is not searchable because it has no real understanding of what it is saying. He adds that tools like ChatGPT can cause other problems for search engines by flooding the Internet with AI-generated, search engine-optimized text. “All search engines will have problems,” he says.
Alex Ratner, an assistant professor at the University of Washington and co-founder of Snorkel AI, which works to train artificial intelligence models more effectively, calls ChatGPT a “legitimate breakthrough” in what software can do. But he also says it may take some time to figure out how to prevent language models like GPT from being invented. Finding a way to keep them abreast of new information to keep the search fresh will likely involve new approaches to training the underlying AI models, he believes.
How long it will take to invent and test these fixes is unclear. It may be a while before the technology can radically change the way people search for answers, even if other uses, such as coming up with new recipes or becoming a study or programming buddy, occur. “It’s amazing, and I told my team that people will see years before and years after ChatGPT,” says Moveworks’ Chen. “But whether it will replace search is another question.”
