The fall of centralized data and the future of LLMs

Gregory Allen, Co-Founder and CEO at Datasent, gave this presentation at our Generative AI Summit in Austin in 2024.

I sold my last company in ’21 and haven’t been doing too much since. Kind of sitting around, trying to make an impact at the acquiring company and not being passive.

But, during that time, I was hankering around and playing with distributed storage, and I was always frustrated with the lack of data standardization and my inability to import data.

And I started working on a new solution.

The issue with bad training data

The problem that I see isn’t really with the AI or necessarily with the companies that are producing some of these language models, but it’s really with the underlying data itself, and the way it’s being centralized and the impact that bad training data can have on the output of any AI model, whether it be generative AI, some ensemble model, binary classifier, whatever have you. If you’ve got bad training data, you’re predicting to get a catastrophic impact.

And I’m worried about people on my trading desk, hedge funds, working with ChatGPT, and making massive trends in three markets, which means very seamless training in general.

And the direction that that can go for making health care decisions, not necessarily for diagnostic reasons, but maybe for determining quality health care.

These are huge problems that can have serious impacts.

The vision for decentralized AI and data storage

My background is in applied mathematics. And, beyond what I’ve already said, I want to discuss the idea of decentralized data storage and training of AI, federated AI, federated ML, and centralized AI, as opposed to what we see now, which is massive, large, centralized, expensive data storage that I personally believe won’t scale.

And I have been trying to prove that mathematically.

I imported my math equation into a generative AI model to see if we could produce an image. We’re not quite there yet, and I don’t think we’re near it; we have a long way to go before AI takes over the world.

But we are right there when it comes to having AI help the existing workforce either make really important decisions or replace jobs like in customer service.