Keeping AI Coding Engines on the Cutting Edge

As a teenager growing up in Chicago, Jeff Wang dreamed of directing Hollywood action movies. As head of business at Codeium, an AI coding assistant startup, he’s now choreographing a different kind of action: He’s the glue holding the company’s finance, operations and sales departments together.

“They all have to interconnect, and I’m just making sure that all the right information is getting to the right place,” Wang told The Forecast in a video interview recorded in May 2024 at Nutanix’s .NEXT conference in Barcelona.

He sees AI innovation moving rapidly, reaching limits then expanding with new possibilities.

“They're still living in the past if they think that we hit a wall because there's so many more modalities where this technology is going,” he said.

Wang has a knack for immersing himself in some of the most buzzed-about technologies in the world. Known for his insights into the crypto markets through the educational platform RocketFuel, he is now firmly planted in the generative AI space.

He first test drove generative AI when he took GPT-2, the precursor to the headline-grabbing ChatGPT, for a spin. He also tinkered with Stable Diffusion, a generative AI model that produces images. But it was ChatGPT that really made him sit up and take notice. Since it started making waves late in 2022, it has turned on their head enterprises in nearly every industry, from education and entertainment to medicine and manufacturing.

As floored as he was by generative AI, however, Wang was convinced that its best capabilities had not yet been seen. That was in early 2023. Soon thereafter, in May of the same year, he joined Codeium to get in on the AI action firsthand. 

Transcript (edited for readability):

Text: The rapid evolution of generative AI in coding.

Jeff Wang: ChatGPT proved, hey, this is possible. There are some really cool things you can do with AI. And at the coding level, people initially were skeptical. They probably wouldn't even believe these things were possible over a year ago. And as they were typing code and it's completing, a year ago again, they're like, hey, it's giving me results. They're okay, right? But the technology is moving so fast that they finally see that trend line. They see that it's going this way, like way up and to the right. And now they're asking for things that are not even possible yet. So a year ago, they didn't believe it, and it was okay. Now it's working very well, and they want more. Everybody is just catching up to today. And I think the customers that saw us a year ago will follow us on this journey. The new customers are going to be like, oh, I think this is possible. I think this is possible. But we're going to make sure everyone's on the same page of where we are today and where we're going tomorrow.

Text: Ensuring ROI and security in generative AI solutions.

Jeff Wang: Everybody has decided that generative AI is here, so let's get some solution in place. One big challenge for us is making sure that they know the ROI and the business value. And maybe if you look at a bunch of other generative AI solutions, it's actually unclear. With coding, it's very, very clear. There's a very high ROI, but then there's a problem with legal and security, and how do you make sure your IP doesn't escape into the internet? So then we take all these points from our customers and come up with a solution that works for everybody.

Text: Building for today while anticipating tomorrow.

Jeff Wang: With every technology, there's always a ceiling that is not really known initially. So when you look at the gold rush, the ceiling is kind of like the supply of gold, right? There's just not much left, but everyone's coming anyway. With cloud computing, actually, I think cloud computing actually did transform a lot of the application world. Like, there's less database engineers now. People can easily go in and manage different tables and deploy apps very quickly. So that ceiling actually does feel like we've kind of hit it, but it took a long time to hit that ceiling. I think AI is actually so early that people, and then people think the ceiling is like, you know, Skynet and the world's over. But there's going to be something in between. And I think it's up to us, the people that are in the industry that are building AI products, to find out what is logically the right place, the right ceiling that we can hit today and give you a product that is utilizing the available technology. And then we're trying to build, obviously, towards things that are not out yet. We're assuming, for example, let's say GPT-5 comes out maybe next year. What are the capabilities, and what should we plan on for our products to be kind of aligned with those capabilities? Or if there's a very large breakthrough in training different models, how do we align our product to actually fit into that mold? And I think a lot of people are in the opposite camp. They see a model get released, and they say, how do we use this model that's released and run it on our Nutanix box? But we have to be more forward-looking than that. If we do what everybody else is doing, we're probably not going to succeed.

Text: The untapped potential of multimodal AI.

Jeff Wang: The news cycle says a lot of things that people want to hear. They want to make sure technology is like: “Oh, there's something wrong. You should read our article.” But there is some truth that. The entire world's data has essentially been trained already. So there is a limitation of how much data has been already put into these models. But there's now innovations on, hey, it's not just text LLMs. Now there's a concept called LMM, like a large multimodal model, meaning there's visuals, there's sound, and then you're combining all these factors into a model. And that has just begun. So people are still living in the past on text. Yeah, we've trained all the text, but there is so much more data on visuals and sound and generating music and generating video. These things are just beginning. And I think people are underestimating. They're still living in the past if they think that we hit a wall because there's so many more modalities where this technology is going.