AI: What is the key to broad adoption?
I had the pleasure of joining Jill Malandrino on Nasdaq TradeTalks recently to talk up our artificial intelligence (AI) technologies.
We dove into why AI at the edge will be essential for AI to scale, and checked out my in-studio demonstration of Stable Diffusion, a text-to-image generative AI model, running completely on an Android phone.
Editor’s Note: The following interview was transcribed and edited for clarity.
Jill Malandrino (JM): Welcome to Nasdaq TradeTalks. I'm Jill Malandrino, Global Markets Reporter at Nasdaq. Joining us at the Marketside Studio in Times Square, New York City, we have Don McGuire, who's a Senior Vice President and Chief Marketing Officer at Nasdaq-listed Qualcomm (NASDAQ: QCOM), also a constituent of our NASDAQ 100 Flagship index. It is great to have you with us, Don. Welcome to TradeTalks.
Don McGuire (DM): Thank you, it's great to be here.
JM: Tell me about Qualcomm; why it is going to be key to the broad adoption of AI.
DM: Actually, AI is present today across billions of devices that have our Snapdragon platforms inside of them — creating amazing experiences like making your pictures sharper or making your sound quality crisper when you're listening to music, or even finding the best connectivity source for you whether it's Wi-Fi or cellular. AI has already proliferated multiple device categories, including smartphones and PCs, because they're powered by Snapdragon.
We're already there in the on-device AI world today, and with this new phenomenon called generative AI — which is all the buzz right now — we have a really, really large role to play.
JM: Why does data need to be distributed between the cloud and the device in order for it to become a part of everyday use? What role does Qualcomm play here?
DM: It's really important that when you look at things like large language models — such as ChatGPT, for example — that as it scales, or in order for it to scale, you need to have a distributed AI environment. Not everything can go to the cloud and then come back because:
- The data center can't support that many instances or that many queries; the cost or economic model doesn't work.
- The amount of power that it takes off the grid also is going to be a problem, so you have to be able to distribute the AI across different parts — from the device at the edge … to the cloud — so that you can do tasks in a very power-efficient way and that AI can then scale more broadly.
JM: And you have a demo for us. Tell us what that’s about.
DM: I do have a demo. This is Stable Diffusion, which is a large generative AI model. It's focusing on taking text and turning it into images, or photorealistic images, and this is running completely on device. There is no connectivity; this is not going to go to the cloud and then come back. This [model] is pre-loaded with over a billion [parameters], so if you put some text in here, in less than 15 to 20 seconds it's going to deliver back a photorealistic image of whatever you asked it to do.
JM: Let’s do it.
DM: Alright, so what do you think we should put in here?
JM: Times Square.
DM: Alright, so we'll put Times Square [types “Times Square” on smartphone] So I just entered Times Square. I'm going to hit generate and you'll see it actually start to render the image as the little clock ticks up — you can see it now trying to pull in Times Square — and then in less than 20 seconds, you will get a photorealistic image of Times Square.
JM: Amazing.
DM: And this is all happening on device; again no cloud connectivity is needed. What’s really important about that too is privacy and security. So, if your data stays on your device, you know that nobody else has it, and with things going back and forth from the cloud, you're not sure if that privacy and security is actually going to be there for you.
JM: Alright, appreciate the insight. Thanks for joining us Don. And thanks for joining me on TradeTalks. I'm Jill Malandrino, Global Market Reporter at Nasdaq.
