Generative AI is taking over startup pitches.
“Nine out of the 10 meetings that I’m taking right now has something to do with generative AI,” said Yuval Neeman, managing director at Trilogy Equity Partners. “But every demo I see is actually super impressive.”
That was one takeaway from a Seattle Tech Week panel discussion Monday called “Supercharging Your Startup with AI,” hosted by venture firms Trilogy Equity Partners and Unlock Venture Partners.
The panelists included: Seasalt.AI CEO and co-founder Xuchen Yao; BrightCanary co-founder and CEO Karl Stillner; Falkon co-founder and CTO Aakash Kambuj; and Alongside co-founder and head of engineering Anish Mehta. It was moderated by Unlock Venture Partners partner Andy Liu and Neeman.
Read on for takeaways from the discussion. Comments are edited for clarity and brevity.
How are you using AI tools internally?
Yao: “A lot of information is carried out in daily conversations and meetings, and you simply cannot participate in multiple meetings at the same time. Nobody has time to jot down everything. Internally, we built our own tool to record, transcribe, and summarize every meeting we have to become an enterprise knowledge base.”
Kambuj: “GitHub CoPilot is bloody simple to use. Every single engineer in this room should be making sure they’re using that. There’s no excuse not to. Our data science team thinks that it’s been a 3X force multiplier for them.”
“In data science, there are all these interesting questions you want answered, but the ROI on the amount of time somebody would need to go and spend in order to find an answer was just not there. It just took too much time to do all that analysis manually. With these algorithms, it unlocks your ability to answer those questions within hours instead of weeks.”
What are some of the hurdles with using AI?
Stillner: “We’ve had hallucinations this past week, where ChatGPT-4 was giving out email addresses for customer service aliases that don’t exist in our company, which is frightening. We basically took that learning and realized that anytime GPT-4 is directly interacting with a consumer, we have eyes on it because it just comes up with crazy stuff occasionally. It’s not that common, but it happens.”
How can a startup using OpenAI’s API differentiate?
Neeman: “The question I ask as an investor is, ‘where’s my edge in selecting who to invest in?’ We’ve landed on two things: One is if the founder has access to a proprietary dataset that no one else has access to. The second piece would be some unique distribution channel.”
What do you see in five years?
Yao: “People always underestimate what’s going to happen in the next five years and overestimate what’s going to happen in the next two years. In five years, robotics are going to take off. Once real-time interaction with robots becomes possible and speech recognition is incorporated, robots will be able to respond within a second. As a result, automated calls, like spam calls, using advanced robotic technology, or ‘Autobots,’ will likely become popular. I believe physical robotics and mechanical robotics will really take off.”
“Another prediction: considering how ChatGPT can generate content ten times better and faster than humans, and looking at the entire history of the internet, which spans less than 50 years, it is likely that deep fake content will be ten times more prevalent than anything produced by humans in the next five years. I think Google or Bing will have a hard time differentiating the true content from the fake content.”
Kambuj: “Spotify recommends playlists for you. What if you could have music that was made just for you? It’s only available to you in your Spotify account, and nobody else has that music based on your tastes. There’s no reason something like that can’t exist. They just made a whole South Park episode that was completely machine-made. Music is coming next.”
