The new shopping guides highlight the potential of generative AI in e-commerce, creating guides for product categories that are too niche to get the mainstream treatment. “The ultimate hedge shears” for example.
Guide supplies
However, the guides also show how generative AI threatens to upend the economics of search and shopping while borrowing heavily from conventional publishers.
Concerns of this nature are unlikely to slow interest in AI at Amazon or any other e-commerce store. Machine learning is already widely used in e-commerce for analytics, search and product recommendation. With LLMs discovering new use cases, one analyst report suggests the e-commerce AI market will grow from $6.6 billion in value in 2023. to $22.6 billion by 2032
“LLM agents are changing the rules of customer service,” says Mark Crystal, CEO of Profitmind, a company that uses AI to provide retailers with analytics.
Crystal says big players like Amazon stand to benefit the most from the rise of generative AI because they have so much data to feed into their models. This should “lead to increasingly capable AI systems that not only improve customer service but also drive product and delivery innovation,” he says, although he notes that “essentially data-rich will continue to become richer and the data poor will become poorer.”
Amazon says its Rufus LLM already demonstrates some unique capabilities that are particularly useful for e-commerce. Chilimbi recounts an incident with an Amazon executive who asked LLM to recommend the best Batman graphic novels and was surprised when he came back with a list that included the non-Batman dystopian classic Guards. When asked why he chose the book, model Rufus stated that the themes and characters in Frank Miller’s popular Batman series from the 1980s The Dark Knight is back carry a similar resonance as those in Alan Moore Guards. “Every once in a while you’re like, ‘Oh wow, how does he do that?'” Chilimbi says.
Amazon’s Rufus LLM not only eats a different diet than most LLMs; it also gets a different kind of fine-tuning. Additional training, which typically helps chatbots engage in coherent conversation and avoid saying inappropriate things, is being used by Amazon to train its model to be a better “shopping concierge.” “There are multiple signals” fed to the model as fine-tuning, Chilimbi says, including whether someone clicks on the recommendation, adds it to their cart, and ultimately buys it.
Chilimbi adds that Amazon has developed its own shopping benchmark to test Rufus and help it get smarter. But while a conventional LLM might be tested on its ability to answer general knowledge questions or solve math or science 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 could eventually enable its independent, shopping-centric AI agents.
“We’re not quite there,” says CMU’s Salakhutdinov, who notes that he still wouldn’t be comfortable giving his credit card to an AI agent. “There are some actions you really can’t give up,” he says. “You know, like you already bought it.”
