Canfield showed WIRED shopping guides for televisions and earbuds that included key technical features, explanations of key terminology and, of course, recommendations on which products to buy. The underlying LLM has access to the vast corpus of product information, customer queries, reviews and feedback, and user purchasing behavior. “This is really only possible with generative AI,” says Canfield.
The new shopping guides highlight the potential of generative AI in e-commerce, creating guides for product categories that are too niche to normally get the treatment. 'The ultimate hedge trimmers' for example.
Guide supplies
However, the guides also show how generative AI threatens to disrupt the economics of search and shopping, while borrowing liberally from conventional publishers.
AI-generated search results now often offer product comparisons and opinions. This diverts traffic from outlets, like WIRED, which make money by producing shopping guides, reviews, and other articles, even though the AI results are primarily produced using data collected from such websites.
Canfield declines to say what additional training data was used to build the new AI shopping guide feature. (WIRED's parent company, Condé Nast, partnered with OpenAI, the company behind ChatGPT, in August this year.)
These types of concerns are unlikely to slow interest in AI at Amazon or any other e-commerce channel. Machine learning is already widely used in e-commerce for analytics, search and product recommendation. As LLMs open up new use cases, an analyst report suggests the market for AI in e-commerce will grow from $6.6 billion in value in 2023 to $22.6 billion in 2032.
“LLM agents are a game changer in customer service,” said Mark Chrystal, CEO of Profitmind, a company that uses AI to provide analytics to retailers.
Chrystal says that big players like Amazon may benefit the most from the rise of generative AI because they can add so much data to their models. This should “lead to increasingly capable AI systems that not only improve customer service but also lead to product and delivery innovations,” he says, although he notes that “essentially the data rich will continue to get richer and the data poor.” will become poorer.”
Amazon says its Rufus LLM already demonstrates some unique skills that are especially useful for e-commerce. Chilimbi recounts an incident involving an Amazon executive who asked the LLM to recommend the best Batman graphic novels, and was surprised when he came back with a list of the non-Batman dystopian classic Watchmen. When asked why it chose the book, the Rufus Model stated that the themes and characters in Frank Miller's popular 1980s Batman series The dark knight returns have a similar resonance to Alan Moore's Watchmen. “Every now and then you say, 'Oh wow, how does it do this?'” Chilimbi says.
Amazon's Rufus LLM is not only fed a different diet than most LLMs; it also takes on a different kind of sophistication. The additional training that normally helps chatbots have a coherent conversation and avoid anything inappropriate is being used by Amazon to train its model to be a better “store concierge.” “There are multiple signals” that are passed to the model as refinement, Chilimbi says, including whether someone clicks on a recommendation, adds it to their cart and ultimately purchases it.
Chilimbi adds that Amazon has developed its own shopping benchmark to test Rufus and help it become smarter. But while a conventional LLM can be tested on its ability to answer general knowledge questions or solve mathematical or scientific problems, Amazon's benchmark tests the model's ability to help a customer find what they're looking for more easily.
Amazon hopes that increasing the shopping IQ of its AI will eventually enable its independent, shopping-focused AI agents.
“We're not quite there yet,” says CMU's Salakhutdinov, who notes that he wouldn't feel comfortable giving an AI agent his credit card yet. “There are certain actions you can't really get rid of,” he says. “You know, like you already bought it.”
