You may have heard the term “backbone of AI.” What does that mean? What is the “backbone” of an AI system?
I found this from Copilot.
“In modern artificial intelligence, the backbone typically refers to a core neural network architecture responsible for extracting meaningful features from raw input data. This concept is particularly common in computer vision, where the backbone is a pre-trained convolutional neural network (CNN) used as a feature extractor for more complex tasks such as object detection, segmentation, and image captioning.”
This is one way to explain it, and I’ve often said that convolutional neural networks are at the heart of AI. Convolutional neural networks allow entities to “see” what is around them and begin processing environmental data, making deployed AI orders of magnitude more powerful. I don’t know if you can call it a spine.
This is from a writer who calls himself “Pete the Momentum” Coreweave system identification It acts as the “backbone” for the AI operator, but we don’t know if it’s just a puff piece or something else.
I found this to be a bit more helpful. Data center knowledge pieces Ivo Ivanov of Industry Perspectives, citing an MIT publication, writes:
“According to the announcement in 2024, MIT Technology Review Research shows that an astonishing 95% of companies are already leveraging AI in some way, and more than half aim for full-scale integration in the next two years. The momentum of AI is nothing short of amazing, but like any emerging technology, there will be peaks and troughs before reaching a state of blissful equilibrium. ”
Boston discussion at Imagination in Action
And this is from an April event at MIT by Imagination in Action, an organization I’m a part of. My colleague Dave Blundin interviewed Jeremy Kepner, director of the MIT Lincoln Laboratory Supercomputing Center, Libby Wayman of Breakthrough Energy Ventures, and Chase Lockmiller of Crusoe via conference call.
energy equation
“I think AI is changing the amount of power needed to run global computing infrastructure by orders of magnitude,” Lockmiller said, detailing plans for data centers across the country, including in Northern Virginia and Denver, Colorado. “It creates a tremendous opportunity to generate not only a wealth of intelligence but also energy.”
Keep in mind that, as he mentioned, Mr. Lockmiller’s company is involved in the Abilene, Texas, plant that is part of the Operation Stargate project, which has a staggering $50 billion allocation announced at the White House.
Wayman called the AI arms race “a U.S.-China phenomenon” and talked about some of the challenges involved in getting to where we want to be.
“In the United States, by the end of 2024, about 4.4 percent of the U.S. electricity consumption went to data centers. So if you zoom out from the specific data point that Mr. Chase shared, the trajectory that we’re on is that data centers could consume about 20 percent of the U.S. electricity, which would essentially lead to a 20 percent expansion of the U.S. power grid and electricity consumption by the end of this decade,” she said. “This constraint that we’re feeling now is actually due to increases of just a few percentage points, and accepting up to that 20 percent increase would put a huge strain on the system. What that looks like in individual local economies is that we’re starting to see models that show electricity prices going up and emissions going up.”
MIT contribution
“I think it was around 2002 that we observed the importance of energy in computing, and it was clear to us here at MIT that energy was going to be very important,” Kepner said.
He talked about building a “next generation” data center next to a hydroelectric power plant in the former industrial town of Holyoke, Massachusetts.
“We wanted it to be not only the best fit for computing, but also for the long term,” he continued. “We knew that no one was going to replace these products for us, and we knew that technology was going to change and this GPU technology was going to change, so we designed a very flexible data center to accommodate basically any generation of technology.”
language and progress
Kepner said two other interesting things. The first was about the supercomputing industry, where most people think progress is rapid.
“People often ask me, has this changed much over the years?” he said. “And I’ve been in supercomputing for a very long time, and as far as we’re concerned, nothing has changed. These computers perform the exact same mathematical operation they’ve been doing for 40 years, called matrix multiplication. That’s the only operation we know how to accelerate. We don’t know how to accelerate anything else. And in fact, AI If we don’t have access to that computation, we’re not talking about it. And that’s been the fundamental driver of AI for 40 years.
interesting.
Another point is where investors can focus their support and efforts, as Kepner urged us to “bet on the math.”
“Mathematics is as much a human language as English, French, etc., and AI understands mathematics much better than English,” he said.
make things go smoothly
Towards the end of his talk, Mr. Lockmiller gave a very detailed perspective on how management is done when it comes to these cutting-edge projects, as speed becomes a top priority.
“Libby mentioned the speed of things,” he said. “You know, when we were building the early stages of Abilene, I committed to getting it done within a year, and we went through this process of trying to order different things from vendors.”
He cited something called a power distribution center, which uses medium-voltage power and distributes it to low-voltage transformers.
“That lead time was over 100 weeks,” Lockmiller said, “and this was from multiple different vendors. And I said, ‘Okay, well, that’s not going to work for me, because I said we’d get this done in a year.’ We actually had a facility that manufactured some of the components. And I turned to the team and said, ‘How fast can we make this?’ And in the end, we were able to complete it in 22 weeks. ”
This is just a little bit about what happens at the top of these big projects. Please look forward to future updates.
