Good morning.
Yesterday, I shared a Fortune report which explained that some experts think generative A.I. is in a “hype curve.” That stirred up some reader responses. Today, I’m continuing the conversation.
“What’s going on technically and organizationally isn’t ‘hype,’” Michael Schrage, a research fellow at the MIT Sloan School Initiative on the Digital Economy, wrote to me in an email, in response to the column.
I’ve previously spoken with Schrage about the intersection of generative A.I. (large language models, or LLMs, like ChatGPT) and the finance function. In January, he also participated in Fortune's CFO Collaborative dinner in partnership with CFO Daily sponsor Workday and Deloitte. So I had a chat with Schrage yesterday, for an update on what he’s observing in his interaction with companies.
“At this time, July 2023, I am comfortable anticipating that we’ll see at least a doubling of LLM-enabled efficiencies and effectiveness within the next 12–18 months,” he tells me.
An example? Let’s say a company wants to embark on an LLM process, Schrage explains. “Whatever your productivity prediction is now—we’ll be able to improve by X percent or save Y amount of money—I believe you'll be able to double whatever your measure of effectiveness is,” he says. “As these models become more powerful, you can figure out how to get more value.”
He continues, "There will likely be a 10x order of magnitude improvement in processes and workflows within five years. I think my estimates are conservative, not bold.”
Improvement in processes is where software companies will come in, Schrage says. “LangChain is a framework that allows developers to link software modalities, programs, and databases to large language models,” he explains. “Increasingly, we're going to see not just use cases, but larger workflows and processes that will require more computational and software resources.” Since LLMs “aren't very good at certain kinds of computation,” LangChain will allow organizations to “get more value in more ways from their LLM investment connection resource,” he says.
But Schrage's prediction on LLM usage over the next few years might not be the case for every organization. Julie Sweet, CEO of Accenture, recently told Fortune CEO Alan Murray what chief executives are saying. "There is a broad and widespread interest, and a reasonable, healthy sense of skepticism, not about whether generative A.I. will be transformative in the long term, but skepticism in the short term about how fast to go," Sweet said. Adding, "we say that cloud is the enabler, data is the driver, and A.I. is the differentiator. Many companies are still building their data core. And to start doing a bunch of experimentation with generative A.I. may not make sense for them."
Although Schrage maintains that generative A.I.'s potential is real and not hype, “I do agree that the FOMO [fear of missing out] issues are not trivial,” he says. “There are scenarios where management and technologists identify and invest in ‘use cases’ that don’t or won’t make economic sense over time for the enterprise.”
Organizations shouldn’t simply say, "‘Okay, we have an existing process, how can we have LLMs automate it or augment it differently?'” he explains. “It’s not just an automatic swap out. It's really being more intentional about: How do we maximize the value from our LLM exploration and experimentation?”
Speaking of generative A.I., a new report by Similarweb, a web analytics firm, found that worldwide desktop and mobile web traffic to the ChatGPT website dropped 9.7% from May to June, and in the U.S., the month-over-month decline was 10.3%. The research also found that worldwide unique visitors to the website dropped 5.7%. OpenAI launched ChatGPT in November.
“I would say 9.7% is a big drop,” Schrage says. “But were those just people coming to the website to just kick the tire?”
I reached out to Similarweb to ask if there was also a drop in returning users as well. David F. Carr, a senior insights manager at Similarweb, shared in an email data that showed there was a smaller drop in returning users than unique users. And there are approximately 21.87 million new users, compared to 47.92 million returning users, according to the findings. “So yes, that might mean those who continue to find value are returning but the supply of new visitors is dropping,” Carr wrote.
Have a good weekend.
Sheryl Estrada
sheryl.estrada@fortune.com
