OpenAI CEO: AI needs a lotta chips, just maybe not $7T worth

OpenAI CEO: AI needs a lotta chips, just maybe not $7T worth

OpenAI CEO: AI needs a lotta chips, just maybe not $7T worth PlatoBlockchain Data Intelligence. Vertical Search. Ai.

FDC OpenAI CEO Sam Altman has driven a lot of hype around generative AI, but as he reminded attendees at Intel’s Foundry Direct Connect (FDC) event Wednesday, not everything you read on the internet is true.

Speaking during a fireside chat with Intel CEO Pat Gelsinger, Altman dismissed reports that he was seeking to raise $5-7 trillion to build a network of chip fabs and the infrastructure to support it.

“If I had to sit there and correct every mistaken report, I would not be able to do my job, but people say all sorts of things,” he said, when teased by Gelsinger about the figure.

The report, published in a Wall Street Journal article citing unnamed sources, understandably attracted a lot of attention. However, many, including this humble vulture, were quick to point out just how big $7 trillion dollars is compared to the entirety of the semiconductor industry.

While Altman’s AI ambitions may not require such a lofty sum, he emphasized that it’s probably a lot more than people expect. “I think everybody is underestimating the need for a lot of AI computing,” he said.

Just how many fabs that translates into, Altman hesitated to say.

“I expect the market for AI compute to be a different kind of good. It’s more like energy, where there’s a certain amount of demand at some price, and less at a higher price,” he explained.

The argument appears to be that with enough capacity, the cost of AI can be made “super cheap,” but getting to this point will require massive investment, not just in fabs, accelerators, datacenters to house them, but clean energy to run them.

“Certainly the most expensive of all, there’s the energy it’s going to take to power those,” he said.

Altman’s emphasis on power isn’t all that surprising considering his investments in alternative energy sources, including both fission and fusion nuclear reactor tech.

In addition to requiring a massive amount of compute and power, Altman also warned that the future of AI development will be rather messy, and we shouldn’t rush into it too quickly.

“There’s obviously a huge amount of understandable anxiety, but exactly what people are going to do, what the right approach is going to be… I’m happy it’s happening now,” he said.

Altman explained that this has informed OpenAI’s approach to AI development, pointing to the company’s “iterative deployment” strategy.

“What you shouldn’t want, I think, is OpenAI to go build artificial general intelligence in secret in a basement and then push a button, like drop it on the world all at once,” he said. “What you want is a very embarrassing GPT-4, slightly less embarrassing GPT-5, pretty good GPT-6, GPT-7.”

He argued that this approach gives time to adjust to AI’s evolving capabilities and undercover potential pitfalls.

And according to Altman, there are no shortage of pitfalls to uncover.

“There are all the sci-fi risks. I think those are actually really super important,” he said. “You can imagine what a tool like GPT-plus-plus-plus can do in the hands of someone who wants to cause harm. And you can imagine the unintended consequences of a system like that. People have been making movies about it for a long time.”

In the short term, Altman predicts implications for elections, cybersecurity, bioterrorism, all which he says will require careful consideration to mitigate. This, he adds, is something that governments and institutions need to be involved in. In fact, a recent report with several OpenAI contributors offered suggestions on just that.

But while Altman was open about the potential harm AI could do, he still believes the tech will ultimately be a force for good.

“When we zoom out and look at if we think the future is going to be a lot better or a lot worse on the whole, I think it’s going to be a lot better,” he said. “In fact, I think it’s hard to imagine how much better it’s going to be.”

As to whether the workers replaced by AI will agree with that sentiment, we somehow doubt it. ®

Time Stamp:

More from The Register