AMD predicts that a full AI agent will be possible "in 3-5 years."

AMD predicts that a full AI agent will be possible "in 3-5 years."

Computex2024, sometimes called AI-tex at this point in time, is undergoing a massive software upgrade, with new chips from Intel, AMD and Qualcomm improving AI performance and taking advantage of that extra headroom (such as Nvidia's Project G-Assist).

However, there is one important goal that everyone is aiming for — a fully autonomous agent AI that can assist you in all of your daily tasks and take action on your behalf, constantly learning and improving yourself.

After talking to Donny Woligroski, Senior Technical Marketing Manager at AMD, we know that the industry predicts this will "become a reality in 3-5 years.""

AI agents are already a kind of thing, but in a very limited way. For example, an email spam filter is an agent acting on your behalf, or an in-game AI that takes action to assist or compete with you.

What we're talking about here is something deeper. Something more similar to an assistant, as you see in the movie "She" (without copying Scarlet Johannson's voice, of course), you can learn to help with all your computing tasks. 

"What if I have a laptop and want to project it wirelessly on my TV right now?" Well, I'll probably google how to do it and probably get some instructions. Then I go into system settings, set these things and then turn it on.Woligroski explained. 

But with a full-fledged AI assistant, it's a different story. "The idea is to ask the laptop "wirelessly project it onto my TV in front of me.""And that would do it."

So far, there are some good multimodal options like GPT-4o and Gemini Ultra, but these are not all like full agents like this. That's the next evolution, and the whole industry is "working towards the end goal" in Woligroski's own words.

"I've never seen the industry move together in a lock step. Like it's really, really out of the box," commented Woligroski. "It means that everyone is looking at the possibility, and everyone is going to move towards that possibility," he said."

It shows that everyone has found something big here, and every company to become the best assistant possible for you, you it will be huge for the accessibility to technology too, or as Woligroski puts it,""It's a non-technophile.""

The Rabbit R1 promised something like this with a massive action model (LAM), but honestly, it didn't turn out to be anywhere near that promise (at least). But Donnie had something to say about this when asking him a simple question: When?

"The most exciting thing about this whole thing is that if you watched an episode of Star Trek 10 years ago where Captain Picard talked to a computer and said, "I assume this thing for me," it was pure fantasy.Woligroski said, have become good and geek with me. 

But now that LLMs is so many things and multimodal AI has become a topic in town, we're moving at a breakneck pace towards that. "So, when does it happen? That is, if this pace is maintained, I think only 3 to 5 years? I think the hard part is that all providers keep up with the features," predicted Donny.

Mark your calendar and in about half a decade we can see a full-fledged AI agent really become a thing.

Categories