What else can generative AI do?

Read it first on my Substack

Back at the beginning of the year, I highlighted “generic sequence encoding” as one of my areas of interest for the coming months. I recognize this is terrible branding. But the idea has stuck with me, both in my own research and in thinking about the generative AI surge.

One way to think about sequence encoding is in terms of cascades. An area of research going back decades, cascades are a conceptual model of behavior that spreads (think sharing a post on social media). They’ve been useful for understanding phenomena like online virality and the structure of influence.

They’re also valuable for illustrating the power of encoding and modeling sequences of events. Cascades were the focus of my research at Meta — -understanding where they started, why they spread, and how large they would get. In this work, and in the paper that came out of it, I focused on finding new ways to measure influence within cascades by identifying their most important participants.

This kind of approach has clear use cases on social media. Understanding which users are consistently early to viral posts lets us see where trends originate and how they spread organically. But there are all kinds of other contexts in which identifying early adopters can be useful. In music streaming, the users who consistently…