Exploring the Frontier: Insights from the Cerebral Valley Generative AI Conference
Observations on the early stages of a technology paradigm shift and the exciting developments ahead.
Wondering what’s happening at the bleeding edge in Generative AI technology development? Last week I attended a conference, Cerebral Valley, focused on just that. It was an amazing set of leaders in the space sharing their progress and talking about what’s coming. Below are my notes:
Generally, the sentiment is we’re at about 1997 in hype cycle comparison to the internet. However, there’s shared enthusiasm that this technology paradigm could keep climbing in impact and company valuations for a decade or more. “This is the super cycle in technology for our generation” - Sarah Guo, General Partner at Conviction.
Pat Grady, Partner at Sequoia, shared sentiments I agree with. Foundation models are already advanced enough for founders to build market-defining application companies across the economy. Services businesses are likely to see the largest changes in the shortest time, and the potential value creation over the next decade is in the trillions of dollars.
Two interesting companies and creative capabilities I’ll be keeping track of: Suno and Runway.
Suno is a platform for making music with prompts. I’ve been experimenting with it for a few days and it is super fun and easy to use! Co-founder Mikey Shulman said some people are spending all day making and refining music, and others are collaborating in groups and creating songs paired with experiences that build unforgettable memories. The guess was we’re at about GPT-3 stage, with exciting potential but some rawness remaining in the models and process.
Runway is a video creation platform. Co-founder Anastasis Germanidis said their goal is to move from simple prompts creating short videos (right now you can generate 4-second clips) to director-level tools for lighting and perspective, and ultimately the creative capability to deliver a full movie through the platform. He said we’re 1-2 years from a movie comparable to a Hollywood production with a $40 million budget. He also said we’re at about GPT-1 stage in model maturity, and from trying it out, we’re seemingly further away from that reality, but the speed of progress is increasing.
More broadly, there’s acknowledgment that GPT-4 still reigns supreme among Large Language Models, and GPT-5 is believed to be close. But OpenAI is no longer the 800-pound gorilla. There are plenty of great model-building companies, and the app layer applying these models to different use-cases is expanding every day, maturing into the next frontier for outsized impact with this technology. Open source remains well-supported and a variety of open foundation models will continue to proliferate. Smaller parameter systems have a clear role to play by delivering increased control, compute savings, and high trainability for specific tasks.