He’s a familiar face for regular viewers of theCUBE — industry analyst Howie Xu recently hosted a panel during theCUBE’s Supercloud 6. Now, he’s host of the tech and business show “Byte Into the Future” and a CUBE Collective founding member.

Xu (pictured) often talks to founders when he has the opportunity or as a part of theCUBE’s Supercloud event series. But he decided he wanted to do more and make it more systematic, he noted.

“The community, the people outside, they wanted to listen, just like people wanted to hear what we want to talk about it,” Xu said. “They benefit from the knowledge, from the insights. I think it’s a kind of win-win for the founders, for myself and then for the community.”

The CUBE Collective is part of a new approach involving an open model. Participants become part of the theCUBE analyst community where theCUBE shares its platform to bring more knowledge and connections, according to theCUBE Research executive analyst John Furrier.

“We have a bunch of other folks in there, and you’re going to see a lot more about theCUBE Collective,” Furrier said. “Join the team, join the community, be part of the conversation, and contribute and take advantage of the resources of theCUBE and all theCUBE alumni and the network effect that goes with that.”

Xu joined Furrier for a CUBE Conversation during an exclusive interview theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed AI advancements and the future of generative artificial intelligence applications.

AI charts the path ahead

It’s been quite a time in the world of AI, especially given the news that Amazon Web Services Inc. CEO Adam Selipsky will step down and be replaced by longtime AWS executive Matt Garman. OpenAI also announced a new flagship AI model named GPT-4o.

“The amazing part is that you basically have pretty much real-time back and forth,” Xu said. “Behind the scenes, it’s more than just the real-time or not. There is a lot of fundamental things in it. One is the multimodal.”

Before, the process would involve voice into text. That would involve some reasoning, before transitioning back to text, in multiple steps, Xu noted.

“The problem with that is the latency, and if each step has an error, it sort of propagates,” Xu said. “Now, it looks like they’re able to do end-to-end, the multimodal. I’m pretty sure there’s still some hard coding thing in it here and there, but it shows the trend.”

Recently, Xu was talking to one of the best research scientists on this topic. The question that remains is, where are things going from the large language model?

“Is that this end-to-end, kind of the multimodal moving forward?” he mused. “I think he and I are on the same page. In the future, it’s not going to be text [to] text, it’s going to be anything. That anything could be video, audio, going through the transformer model.”

On the other side, it could be audio, video, image or text, according to Xu. While we’re not completely at that point yet, progress is taking place.

“I think that’s where the world is moving toward,” he said.

Here’s theCUBE’s complete video interview with Howie Xu:

Photo: SiliconANGLE