Amr Awadallah of Vectara, Stephen Balaban of Lambda and Erik Bernhardsson of Modal had a lively discussion about AI in the enterprise, with Awadallah specifying that the three big challenges were accuracy, security, and explainability.
Balaban said that understanding how the AI generates its answers, and making that transparent, was critical for many industries. Referencing a discussion with an insurance provider, Balaban said: “We can’t use a black-box model, that’s just not an option. Frankly it will never be an option.”
He questioned whether generative AI in the enterprise would evolve in the way people expect. “There’s this assumption that oh, well, this is all just gonna eventually sort of asymptotically converge to old-school SaaS margins, and it very well may not,” he said. “There’s a lot of players in the stack… running an AI application is very compute intensive, and it’s not at all like serving a web app.”
Bernardsson said that most of the large enterprise firms’ experimentation with AI is coming out of the R&D budgets, but he was still bullish on more wide adoption in the next year or two. “I think it just takes time to find something that works,” he said. Still, he didn’t entirely agree that those 90% SaaS margins would never come, arguing they were achievable on the “serving” side, though it would be much tougher for foundation model companies.
On the topic of lessons from the the past, Awadallah declared that we were indeed in a bubble with insane valuations-he compared it to 2008, when he took a humble $5 million investment on an $8 million pre-money valuation for Cloudera-but then noted that great companies and technologies had emerged from previous bubbles.
Bernardsson pointed to the fiber-optics build-out during the telecom frenzy of the late 1990s as one example.
CEREBRAL VALLEY NEW YORK TOOK PLACE ON JUNE 27, 2024, HOSTED BY NEWCOMER & VOLLEY
cerebralvalley...
www.newcomer.co/
www.volleygame...
Videos made by to www.storycastp...