The Race to Build a ChatGPT-Powered Search Engine

Another problem with a system like ChatGPT is that its answers are only based on the data it was trained on. Completely retraining the model can cost millions of dollars due to its size and the amount of data. YouChat gets confused when asked for the latest sports scores, but knows what the weather is like in New York right now. Socher does not want to reveal how current information is incorporated and sees this as a competitive advantage.

“I think right now a lot of these chat interfaces are vastly superior to the search experience in some respects, but clearly far worse in other respects,” says Socher. “We are working to reduce all of these issues.”

Aravind Srinivas, founder and CEO of search startup Perplexity AI, who previously worked at OpenAI, says the challenge of updating a ChatGPT-like system with up-to-date information means it has to be combined with something else. “They will never be good search engines on their own,” he says.

Saam Motamedi, a venture capitalist at Greylock Partners who has invested in AI-based search firm Neeva, says it’s also unclear how compatible chat interfaces are with search engines’ primary revenue model — advertising. Google and Bing use search queries to select ads that appear at the top of the list of links provided in response. Motamedi suspects that new forms of advertising will need to emerge in order for chat-style search interfaces to work, but it’s not entirely clear what those will be. Neeva charges a subscription fee for unlimited ad-free searches.

The cost of running a model like ChatGPT at the scale of Google could also prove problematic. Luis Ceze, co-founder and CEO of OctoML, a company that helps companies reduce the cost of deploying machine learning algorithms, estimates that it can be 10 times more expensive to perform a ChatGPT search than a Google search, since each answer requires a large and complex AI model.

The extent of ChatGPT mania has surprised some programmers and AI researchers familiar with the underlying technology. The algorithm at the core of the bot, called GPT, was first developed by OpenAI in 2018, and a more powerful version, GPT-2, was unveiled in 2019. It’s a machine learning model designed to take in text and then predict what’s coming next, which OpenAI can impressively demonstrate when trained on huge amounts of text. The first commercial version of the technology, GPT-3, has been available to developers since June 2020 and can do many of the things ChatGPT has been hailed for lately.

ChatGPT uses an improved version of the underlying algorithm, but the biggest leap in its capabilities comes from OpenAI, where humans provide feedback to the system on what constitutes a satisfactory response. But like the text generation systems before it, ChatGPT still tends to reproduce biases from its training data and “hallucinate” plausible but false results.

Gary Marcus, professor emeritus at New York University and vocal critic of the AI ​​hype, believes ChatGPT is unsuitable for searching because it doesn’t really understand what it’s saying. He adds that tools like ChatGPT could create other problems for search companies by flooding the web with AI-generated, search-engine-optimized text. “All search engines will soon have a problem,” he says.

Alex Ratner, an assistant professor at the University of Washington and co-founder of Snorkel AI, which works to train AI models more efficiently, calls ChatGPT “a legitimate twist” on what software can do. But he also says that figuring out how to stop language models like GPT from making things up can take a while. He believes that finding a way to update them with new information to keep the search fresh will most likely involve new approaches to training the underlying AI models.

How long it will take to invent and prove these fixes is unclear. It may be some time before technology can radically change the way people search for answers, even as other use cases emerge, such as: B. coming up with new recipes or serving as a study or programming partner. “It’s amazing, and I’ve told my team that people will be watching ChatGPT for years before and after,” says Moveworks’ Chen. “But whether it will replace searching is another question.”