This week on the GeekWire Podcast, we’re featuring a panel discussion from the recent GeekWire Summit in Seattle, with three technology and business leaders offering first-hand insights into the new era of artificial intelligence.
- Bridget Frey, CTO at Redfin, the tech-powered real estate company that operates in more than 100 markets in the U.S. and Canada.
- Inbal Shani, chief product officer at GitHub, the software development platform used by more than 100 million developers around the world.
- David Shim, CEO at Read AI, a Seattle-based startup using AI to provide a new window into meetings, and transform them in the process.
Frey, Shani and Shim share real-world examples of AI impacting software development, real estate, and meetings. They address topics like privacy, bias, education, and the future of work. They also discuss the changing nature of technical jobs, and a blurring of the line between developers and non-developers.
From the audience, we get questions about adapting AI to account for emotional intelligence; advice for aspiring engineers; preventing synthetic content from corrupting human experiences; protecting proprietary corporate data; and the prospects for improving work-life balance as AI increases productivity.
Listen above, and keep reading for notes and takeaways.
How Redfin is using AI: “It’s been off the charts,” Frey said. “Really, the amount of innovation and the pace of innovation around AI has picked up so much.” The areas where AI has had the most impact are ones that previously involved humans absorbing and synthesizing large amounts of information.
- Using ChatGPT to generate 4x more localized real estate content.
- Developing new AI-powered features across the company in areas like the home valuation process, including explaining home valuations to customers who ask how Redfin arrived at the number.
- Accepting around 25% of code suggestions from GitHub Copilot, increasing efficiency. Copilot is like a virtual AI teammate that operates alongside developers, automatically generating code and helping them do their work.
The average code acceptance rate for GitHub Copilot is even higher, as much as 55%, depending on the use case, customer and license type, Shani said.
Senior engineers tend to have the most success with Copilot, because they are accustomed to explaining things to more junior engineers, and they can adapt that skill set to Copilot, Frey said. However, younger engineers can play an important role in introducing an organization to the cutting-edge tools they’ve been using in school.
GitHub is now focused on applying AI across the software development life cycle, especially on enabling developers to write more secure code, Shani said. This will come through at the upcoming GitHub Universe conference, she said.
The company also takes responsibility for the next generation of developers in its education program, which has more than 5 million learners. It’s preparing the next generation of developers for a world where AI is table stakes, Shani said.
AI product developers need to be wary of information overload, and up front about privacy issues.
- ReadAI initially provided real-time meeting information, telling speakers, for example, whether people were paying attention to them in the moment. Based on user feedback, it shifted to a post-meeting recap instead.
- The startup notifies meeting participants that recording and analysis is occurring at the outset, and allows users to opt out of the meeting and have all related data deleted by typing “opt out” in the chat, Shim said.
Frey on dealing with issues of bias: “Redfin has 20 years of experience with fair housing law, and the large language models have no experience with it. But we can’t simply reencode years of bias into algorithms and call it a day. So we have been partnering with the large language model companies to share with them test cases and rules that we think need to be handled by these large language models, so that they answer questions that in a way that does not violate fair housing law. We’re very optimistic that we’re going to be able to make real progress on this.”
- Another issue, she said, is making sure that AI doesn’t skew toward pleasing people with its answers at the expense of accuracy.
Social norms around AI are still evolving: “I think the most embarrassing thing is when your AI beats you to the meeting, when you’re five minutes late. That kind of makes you look bad,” Shim said. “People have asked us, ‘Hey, can you make it come in a minute or two after I join?’ “
Shim’s advice for AI startups: “You still have to have your own secret sauce. You need a moat. You just can’t go in and say, I’m another wrapper company … someone who uses ChatGPT, LLaMA, Bard, something else, and just does does prompt engineering. You could actually grow pretty big. There are a number of apps on the marketplace right now that do $100,000 or $200,000 a day in terms of sales, but it drops off pretty quickly because it isn’t differentiated.”
- He notes that around 93% of ReadAI’s processing/engineering resources go into proprietary models, and the remaining 7% is focused on outside models.
Shani says we’re in the Industrial Revolution for AI. “Eventually we’ll find a balance between that niche AI and the generalized AI,” she said. “We’re still working in specific areas of domain expertise. There is only so far that generalized AI can take us, and we will need to build on top of that.”
Each panelist’s concluding message to the audience:
- Frey: “There have been parts of our business that we thought technology couldn’t really change. And just in the last few months, as these technologies have come to light … it’s just created this wave of innovation. So I just encourage you to look through your business, look at parts of your business that you thought technology could never touch. Maybe now there is a way that you could make that part of your business better.”
- Shani: “For us, it’s a lot about developer happiness, which comes from developer experience, developer productivity and developer collaboration. It’s about how we continue innovating and taking AI across the software development lifecycle. And it’s a lot of focus on education across the board … to make sure everyone is ready for the revolution that is happening right now.”
- Shim: “Embrace AI. Go as fast as you can. But make sure you have a moat, because this is going to be a period of time where everything is condensed. Things are moving faster today than they ever have.” Iterate quickly, change as needed with the market, but maintain your core areas of differentiation.”
Subscribe to GeekWire in Apple Podcasts, Spotify, or wherever you listen.
Podcast audio editing by Curt Milton. On-site A/V by Adavanza.
