ChatGPT itself can also remember things about users that it can use later. Sometimes it does it automatically, or you can ask it to remember something. These “long-term” memories affect how he responds to discussions. Search doesn’t have this feature yet (a new web search starts from scratch), but it should get the feature in the “next few quarters,” Fry says. When it does, OpenAI says it will allow it to provide much more personalized results based on what it knows.

“These can be persistent memories, like ‘I’m vegetarian,’ or contextual, like ‘I’m going to New York in the next few days,'” Fry says. “If you say, ‘I’m going to New York in four days,’ he can remember that fact and the nuance of that point,” he adds.

To help develop ChatGPT’s web search, OpenAI says it has leveraged its partnerships with news organizations such as Reuters, the Atlantic, The WorldTHE Financial TimesAxel Springer, Condé Nast and Time. However, its results include information not only from these publishers, but also from any other online source that does not actively block its search crawler.

It’s a positive development that ChatGPT is now able to retrieve information from these reputable online sources and generate responses based on them, says Suzan Verberne, professor of natural language processing at Leiden University , who studied information retrieval. It also allows users to ask follow-up questions.

But despite the improved ability to search the web and cross-reference sources, the tool is not immune to the persistent tendency of AI language models to either make it up or get it wrong. When MIT Technology Review After testing the new search function and asking for holiday destination ideas, ChatGPT suggested “luxury European destinations” such as Japan, Dubai, Caribbean Islands, Bali, Seychelles and Thailand. He proposed as a source a article of theTimes, a British newspaper, which listed these locations as well as those in Europe as luxury vacation options.

“Especially when you ask about false facts or events that never happened, the engine may still try to formulate a plausible answer that is not necessarily correct,” Verberne explains. There’s also a risk that misinformation could seep into ChatGPT’s responses from the internet if the company hasn’t sufficiently filtered its sources, she adds.