Leaders from Adobe, Wayfair, Wipro, Freshworks and Medtronic, who met for lunch at Fortune's Brainstorm Tech Conference on Tuesday, said their companies are finding many useful ways to introduce generative AI into their organizations. But many added that the cost and reliability of generative AI tools remain major challenges, and they discussed how they're addressing those issues. Additionally, several noted that choosing the right tool for the right use case is essential.
For example, Fiona Tan, chief technology officer at online retailer Wayfair, said the company is currently working on a second generation of an AI-driven “co-pilot” to assist customer service and sales reps, as well as offering image generation capabilities to help customers envision room designs.
But Tan said the company had devised a matrix to identify whether generative AI would be effective for a given task and how to measure success, which was especially important given the risk of hallucinations — the inaccurate or meaningless outputs that generative AI models sometimes produce.
Some tasks also help customers tolerate illusions better, Mr Tan explained. “They like the fact that even if the chair legs are a bit askew or the things in the room are a bit messy, they can still see the room in different styles,” he said, pointing to the website's interior design ideas feature.
Bedrooms and operating rooms
Meanwhile, Ken Washington, senior vice president and chief technology officer at medical device maker Medtronic, said that although the company uses machine learning extensively, the operating room is not currently the right place for generative AI. “I can't stand hallucinations,” he explained. “That's why we don't have generative AI in medical technology development, and why we don't have any FDA-approved generative AI treatments or diagnostic tools.”
For Adobe, addressing the cost and reliability of generative AI is the same as with any other technology. “We ask ourselves, what problem are we trying to solve? If it's a new problem, that's a red flag,” says Ely Greenfield, CTO of digital media. “Maybe we'll look at a problem that we know already exists, but they just didn't have the right solution before, and this might be the right solution.” And while knowing it won't be 100% accurate and will be expensive, “we take the problem we want to tackle and try to triage early on whether we can determine upfront whether the cost and reliability issues are an issue for the work,” he explained.
But he warned against focusing on costs prematurely. “The dynamics of this world are changing so rapidly that I think the first step is to figure out whether we can use AI to solve the problems we have today,” he said. “By focusing on different models, swapping them out, getting smart people to refine the models, waiting a month, waiting a week, these things will get better. Costs are coming down. So let's work on finding solutions now and worry about costs later.”
But cost is definitely an issue when it comes to adopting generative AI for Freshworks, a SaaS company that provides tools for customer service, IT and sales, said Siddharth Agarwal, Freshworks' svp of product strategy and operations.
“We're a small company, and cost is important to us,” he said. What the company has done is created a “model garden” for its cloud engineering team, making 40 to 60 different models available to developers. Developers can try out the models, determine the cost profile and performance, and determine if it's the right, most cost-effective model. “This allows developers to experiment and optimize on the cost curve,” he explained.
Costs are also a concern for Wipro, an IT consulting firm that hosts a variety of models on its platform, said Chief Technology Officer Subha Tatavarti. Given the cost of computing power, Wipro is looking to smaller models that don't use as much computing power. If Wipro can reduce the cost of offering the models, “it will help our customers reduce costs,” he explained.
Wipro is also focusing on RAG (Search Augmentation Generation), a technique to reduce hallucinations and improve model output. “Our efforts over the past few months have shown very promising results,” she said.
Read more articles from Brainstorm Tech 2024:
Interpublic CEO Philip Krakowski says ad industry needs 'no choice' to adopt AI
From Alphabet's CapitalG to Norwest, how VCs are dealing with the IPO lull: “We're not here to time the market”
Why business leaders see AI as an opportunity to “make work easier”
CEO Daily provides essential context for the news business leaders need to know. More than 125,000 readers trust CEO Daily every weekday morning for insight on and from the C-level. Subscribe now.
