But Salakhutdinov says having plenty of information about how users handle common and significant tasks like shopping can be a key factor in helping them stay on track. “Data will be very important,” he says.
Send it
Chilimbi and Mehta say an agent may eventually go on a shopping spree when a customer says, “I’m going camping, buy me everything I need.” An extreme, but not impossible, scenario would be for agents to decide when a customer needs something, then buy it and ship it to their door. “Maybe you could set a budget,” Chilimbi says with a smile.
Amazon’s fresh AI-generated shopping guides, announced today at the Reinvent conference in Nashville and initially available on the company’s U.S. website and mobile app, are a diminutive step toward the ultimate vision of a super-intelligent shopping assistant. Rufus LLM is designed to automatically generate information and insights that could take someone hours of internet searching to gather. “If you ever try to shop in a category you’re not familiar with, it can take a long time to understand the lay of the land, the different features available, and the different choices,” says Brett Canfield, senior product manager on Amazon’s personalization team.
Canfield showed WIRED buying guides for TVs and earphones, which included significant technical information, explanations of key terms and, of course, recommendations on which products to buy. Basic LLM has access to a huge collection of product information, customer questions, reviews and opinions, and user purchasing habits. “This is only really possible with generative AI,” Canfield says.
Fresh shopping guides highlight the potential of generative AI in e-commerce, creating guides for product categories too niche for normal treatment. For example, “complete hedge trimmer”.
Guiding materials
But the guides also show how generative AI threatens to upend the economics of searching and purchasing while borrowing freely from conventional publishers.
AI-generated search results now often include product comparisons and reviews. This diverts traffic from outlets like WIRED that make money by producing shopping guides, reviews and other articles, even though the AI results are first created using data pulled from such sites.
Canfield wouldn’t say what additional training data was used to build the fresh AI shopping guide feature. (WIRED’s parent company, Condé Nast, started cooperation with OpenAIcompany behind ChatGPT, in August this year.)