I am Ion Stoica. I'm a faculty at UC Berkeley, working at the intersection between AI systems, cloud computing, and a bit of security. I am also a co-founder and executive chairman of Databricks, and at Anyscale I also co-founded and am a board member of ops systems.
The problem is very simple and sits at the intersection of some very powerful trends. On one hand, we have AI in general and generative AI in particular as huge vectors of growth. It's pretty clear that while there is a little bit of hype today, long term they are going to transform and impact every industry. So no enterprise can afford to not at least experiment and use it to enhance their products and services, to improve their businesses, whether it's top line improving revenue or bottom line cutting costs. In order to get value from generative AI, what enterprises have as their edge is their data, not their models, because the models are going to be the same kind of models, be it open source or some of the existing proprietary models like ChatGPT, Anthropic, or Gemini. So on one hand, companies want more and more data to train, and a lot of data is sensitive because sensitive data typically has more value. On the other hand, there are more and more regulations on what you are allowed to use and how you are allowed to use this data. We know that GDPR or California Consumer Privacy Act put strong constraints to enforce the privacy and confidentiality of the data. But when it comes to generative AI, there are more and more new regulations. The last time I looked, only in the US across all the states, there are 700 bills in the world that want to regulate AI in some shape or form. So on one hand, you want to use more and more data, more proprietary data, more valuable data. On the other hand, the constraints to use data are just growing and growing. So how do you bridge this growing gap? That's the question. And the answer to that question is confidential computing. Basically, you want to process this data, get insights, run gen models, but you want to do all of these things on encrypted data. Encrypted data by definition shouldn't leak any information about the data. So it's a general, very powerful solution. And when we say that it's future proof, this is what we mean, because it's going to be robust to new regulations. It's not a point solution which is just good enough for today's world.
6 more exchanges in this transcript
Sign in free to read the rest of this interview. No card required.
