The rapid rise of ChatGPT and other AI-driven creative writing apps has sparked a furious race to build new companies based on the technology. But several experienced Boston entrepreneurs say the best opportunities may lie in adding ChatGPT features to more basic software programs.
“A lot of people are building AI companies,” Brian Halligan, cofounder of HubSpot, said last week at the Imagination in Action conference held at MIT. Instead of joining that crowded field, Halligan told the audience of hundreds of students and startup founders that AI can be used to improve “any piece of enterprise software out there.”
The new AI apps, which can spit out chat answers and explanations that sound remarkably human, could help developers build an easier-to-use front end for apps, he said.
“Every piece of enterprise software is going to go through this transition similar to when the world of DOS went to Windows in the way you interact with the software,” Halligan said.
HubSpot has already built a feature dubbed ChatSpot that can help users find trends in data and build reports, he said. (Other local companies working with ChatGPT include Constant Contact, Drift, and Pegasystems.)
Steve Papa, who sold his company Endeca to Oracle for $1.1 billion in 2011, said he has seen a repeating pattern in the Boston startup ecosystem, as new software companies arise to take advantage of new technologies. “The same companies are rebuilt, they get acquired, and then a whole new generation [emerges],” he said. “For every one of them, AI is going to fuel it, accelerate it, make it more capital efficient, make it grow faster.”
Dileep George, who cofounded Vicarious AI, warned that adding ChatGPT features to software may be tricky, however, given that the program sometimes (and famously) makes factual mistakes. Google’s stock value dropped by $100 billion on one day in February after investors noticed a mistake by its Bard AI search app.
“It’s very easy to create a demo that will blow an investor away,” George said. “But to make it to deployment... there is a big gap in real world deployment.”
The conference, organized by John Werner, also featured a live Zoom appearance by Sam Altman, the chief executive of ChatGPT developer OpenAI. Altman defended his California firm against recent criticism that it was moving too quickly and releasing potentially dangerous AI software.
Last month, more than a thousand academics and experts, including researchers from MIT, Harvard, and Northeastern signed a petition demanding a six-month delay in further development of generative AI systems.
Altman said he agreed with “parts of the thrust” of the letter, explaining that OpenAI had done significant safety testing and auditing on its latest AI model, GPT-4, before making it available to the public.
“I also agree that as capabilities get more and more serious, the safety bar has got to increase,” Altman said.
But the letter is “missing most technical nuance about where we need to pause,” he said. OpenAI is not yet building GPT-5, he added. “We won’t for some time,” he said.
At the end of his Zoom talk, Altman was asked how the audience could know his appearance was real and he was not an AI. “You don’t,” he said and logged off.