Sam Altman Envisions AI Agents as the Next Big Thing

OpenAI’s CEO, Sam Altman, foresees a future where artificial intelligence is marked by the emergence and growing popularity of personal assistants. These will become integral to daily life, potentially even replacing smartphones.

In an interview with MIT Technology Review, Altman described the “killer application” for AI as a “super-competent colleague” that knows everything about a user’s life, every email, and conversation. Yet, the assistant will not “feel like an add-on.”

“What you really want is just something that will help you,” Altman said.

He added that while some tasks could be solved instantly by AI, more complex ones might require further exploration, with the assistant returning for clarifications if needed.

Compared to current OpenAI offerings like DALL-E, Sora, and ChatGPT, which Altman referred to as “incredibly dumb,” this would be a significant leap forward.

The Need for New Hardware

Altman is confident that such a future will not require special devices, as applications will operate in the cloud.

“I don’t think new hardware will be necessary for this,” he stated.

However, the OpenAI chief acknowledged that users would likely “be happy” to see a specialized gadget. At the same time, Altman does not consider himself an expert in hardware.

“I’m very interested in consumer hardware for new technologies. I’m an enthusiast who enjoys it, but it’s far from my area of expertise,” he said.

The Challenge of Training Data

A significant hurdle in creating powerful AI agents is the lack of training data. OpenAI is scouring online archives, newspapers, and blogs to train its models.

However, much of the internet has already been scanned, and access to private databases is challenging. Nonetheless, Altman remains optimistic.

“I believe, but I’m not sure, that we’ll find a way out of this situation […]. Humans are direct evidence of alternative ways [of learning]. And I hope we’ll find them,” Altman opined.

The Race for AGI

Altman suggested that in the race to create artificial general intelligence, several different versions of superintelligence will emerge, each excelling and faltering in various tasks.

“We’ll have to overcome a certain threshold of computational power. But even then, I wouldn’t be so sure,” he added.

When asked if Altman knows the timeline for GPT-5’s release, he briefly replied, “Yes.” However, the startup’s head did not elaborate further.

In April, the LMSYS platform introduced the gpt2-chatbot language model, which has been linked to OpenAI.