Synthesize 2023: Closing Remarks

Ali Golshan (00:26):

Well, everyone, thanks for sticking with us for the entire day. We really hope you enjoyed this first ever synthetic conference for developers. We're excited to say this has been such an enormously-successful and well-received conference by all of you, thanks to all of you, that we actually plan to do this conference in person next year. So that's a really exciting piece of news I'd like to share. And I also want to thank all of our speakers, presenters, startup, partners, and customers who joined us today. It was a great learning experience. I'm really appreciative to be able to get such a deep dive on all things, from privacy to federated learning to some of our partners really talking about some of the cutting-edge work that they are doing. So that's been tremendously exciting for us.

(01:06)
But ultimately, really, what I want to leave you all with is that this is why we are so excited about synthetic data in this entire space. It's because we believe that while there is an entire ecosystem and economy being built for AI that is dependent on that data, that doesn't have to be on a collision course or it doesn't have to cut the other way when it comes to privacy, safety, and even the economic sharing when users come to play. This is where we believe synthetic data, while not a silver bullet, can solve many of the problems that currently are being talked about or potentially being challenged by the traditional means of work. So, this is where we believe users and customers can actually fairly participate in the economics of this particular industry, this is where we believe that the quality or the velocity of data does not need to be compromised when it comes to privacy and safety. Privacy can come first. At the same time, we can actually leverage the technologies that are being developed. This is also why we believe, as I said at the beginning, we do not have to be bound by companies, organizations who build walled gardens around data. Data should be free. Data should be exposed to everyone. And if it isn't, all should be able to participate in the economics of that by actually being able to share it safely with the rest of the ecosystem.

(02:18)
And then finally, we're really excited about some of the fascinating technologies coming out in privacy. It was great hearing Peter talk about how federated learning is potentially creating other solutions in this space that may not be synthetic. So we are very much aware and really wanting to participate in a collective community work that can help us all move forward this particular needle in the generative-AI space.

(02:41)
And then finally, hopefully you got a little bit of a glimpse of all the types of technologies and tools that we plan to bring to the market. Our entire thesis is really what we call open learning. So if you're more interested about what we are doing, we typically release a lot of our initial code as open source so everybody can be exposed to it, validate it, confirm it, and really be able to build your own confidence and trust around it. We tend to write a lot of content around the work we do so you can get a fundamental understanding of how we want to transition that into a productized world.

(03:13)
And I think you probably all got a sense of this, is, especially at Gretel here, we are really focused on enterprises and businesses. And while we think consumer is a really exciting world, and a lot of dependency of that, as we talked about, comes from the public domain data, we are really interested and focused in unlocking and cracking that last mile, being able to leverage that truly sensitive, truly insightful data that can help us go that last mile, create those customized workflows, create those economic participations from companies, without actually compromising safety. And that's really what we are focused on.

(03:47)
So if you're an enterprise customer, user, startup, or just a single researcher who's interested in learning more or experimenting more, please do reach out to us. We would love to hear from you. We always love collaborating. We have a very large applied-science team that works constantly with our customers and users on various papers, operationalizing those papers, or even actually building them as one-offs if we think they can be helpful to the community. But overall, we are here to help. We want to make privacy as a core pillar to this entire industry going forward. So while synthetic data is the right answer now, as you heard from a lot of our other researchers as well as my co-founders, it's not about just training synthetic models, we think a really right approach and a complete approach is a one platform, such as ours, that is multimodal, that can have synthetic models, but can then use those synthetic models and its data as a seed to be able to take those foundation large language models the last mile. We think that's really the intersection that actual economics and value will be created.

(04:45)
And then ultimately, what we talked about way, way back at the beginning of the conference, which is that synthetic data and its nature, by being able to boost or actually combine different data sets, can actually be the answer to differentiating as far as data goes. So rather than, for example, trying to find data that is unique or spending an enormous amount on acquiring data that is potentially unique, synthetic data actually has the ability to be able to create unique insights, or boost unique insights, and be able to create that edge, even if it's for testing or simulation or initial work before the model rolls out, or eventually as a model rolls out and you want to ensure that that model is building anti fragility into its system and is resilient when it comes to real-world conditions.

(05:30)
So, this has really been exciting. We want to thank you all. We really appreciate you all tuning in. We are all going to be available on Discord or Twitter. So if you have any questions, please jump in, ask us. We would love to engage with you, hear the good, the bad, and the ugly. We would also love to hear from you as to what you see to be the challenges or blockers when it comes to synthetic data or generally at getting access to data. Our ultimate goal is making data safe and private so everyone can share and collaborate on it so the data becomes the ultimate equalizer. And it's really that innovation that ends up driving differentiation and value for users and customers. So, on that note, I will end here. Thank you all, and hope to see you all in person next year at _synthesize. Thank you very much, and have a great day.