The smartest machine learning people are in on a secret. They’ve seen how well the algorithms already perform today. They know the technology is ready to move from cool demos to mass-scale, commercial-grade adoption. However, even the practitioners struggle to predict the order in which practical use cases will emerge.
Consider AI in creative fields. People assumed it would take years if not decades for AI to transform art. Dead wrong. Generative models now produce artworks that genuinely impress us and even win awards. But that’s not the most important point.
The game-changer is that large, sophisticated models have become cheap to train. This step change expands the frontier of what software can do further and faster than any time in recent memory.
We’re looking for infrastructure and application-level solutions to problems that couldn’t be tackled or even imagined 3 years ago. A new generation of infra companies will allow any software developer to build ML into their applications without needing a Ph.D. As for end-customers, most of us will use these products like we do electricity: it just works and we don’t have to think about why.