Startups in the early stages of building generative AI tools should concentrate on specialized verticals such as healthcare, biotech, and legal industries.
That’s one key takeaway from a panel discussion earlier this month about the new era of artificial intelligence, hosted by the entrepreneur program TiE Seattle at the Tech Mahindra offices in Redmond, Wash.
“Just be thinking things that big tech companies aren’t thinking about,” said panelist Sabrina Wu, a Madrona Venture Group investor. “There’s tons of opportunity there.”
Companies like Microsoft, Amazon and Google are in a race to integrate large language models into their core business strategies. The tech giants are delivering tools in online search, software development and office software, leaving gaps in specialized use cases. Startups are emerging to fill these gaps, securing hordes of venture capital even as investors have been cautious about deploying capital.
To make sense of the growing adoption and opportunity for startups, tech leaders shared their outlook during a discussion titled “Generative AI: Ushering in a New Digital Era.” The panel included University of Washington computer science professor Ankur Teredesai; Pictory CEO Vikram Chalana; Matt Sinclair, global director of AI/ML sales and strategy at Microsoft; and Wu. The discussion was moderated by Madhu Singh, Foundry Law Group chief legal officer and former TiE Seattle president.
Read on for our three main takeaways and watch the entire discussion in the video below.
Growing adoption
- Two factors have led to the rapid adoption of large language models: accessibility for consumers, as well as the ability to interact with these models using natural language. Microsoft’s Sinclair referenced Andrej Karpathy, a former senior director of artificial intelligence at Tesla, who tweeted in January that the “hottest new programming language is English.”
- Despite a broader funding downturn for startups, Wu said most venture capitalists are thrilled about the opportunities of investing in generative AI startups. She said many venture firms are sitting on piles of dry powder — capital that’s yet to be invested — and expects a steady pace of deployment for startups in the generative AI space.
- At the enterprise level, organizations are training large language models on their own data to offer different use cases to their employees and customers, Sinclair said. These organizations are using the models to get rid of the mundane aspects of their jobs. He pointed to GitHub Copilot as an example, which gives software developers a virtual AI pair programmer to suggest code and functions as they create apps and services.
The challenges and pushback
- The UW’s Teredesai said AI has “shaken academia to its core.” He said educators must balance preparing students to use the latest generative AI technology for productivity, while also raising awareness about its potential pitfalls such as hallucinations. (Previously: UW professors on using ChatGPT in the classroom)
- The panelists agreed that there should not be a moratorium on experiments regarding generative AI. However, they emphasized the need for responsible frameworks and regulatory guidance that can evolve alongside the new technology.
- Generative AI replacing jobs is a growing concern, but Sinclair likened it to horses being replaced by cars, which led to new job opportunities in areas like manufacturing and financing. He said companies should retrain their workforce to use generative AI in their daily work.
How startups can counter position
- Generative AI’s recent advancements have made it cheaper and less time-consuming for startups to use large language models. However, Wu warned that just “wrapping up GPT-3” into a tech stack is not enough. In order to build a successful and defensible company, she said, startups must be solving a novel problem while understanding their target customer needs.
- Wu stressed the importance of user experience for generative AI startups, advising them to create a platform that solves a problem without being intrusive. She cited Gmail’s autocomplete feature as an example of a seamless integration of AI.
