Like many universities, Georgia Tech is grappling with how to provide students with the training they need to prepare for the latest big change in the IT job market: the rise of generative AI (genAI).
Through a partnership with chipmaker Nvidia, Georgia Tech's College of Engineering has built a supercomputer called the AI Makerspace. We use 20 Nvidia HGX H100 servers with 160 Nvidia H100 Tensor Core GPUs (graphics processing units).
These GPUs are powerful. A multiplication operation that would have taken her 50,000 students at the school 22 years to accomplish takes just one second on one of her Nvidia H100 GPUs. So 160 of these GPUs will give students and professors access to advanced genAI, AI, and machine learning creation and training. (This move prompted Georgia Tech to offer her new AI-focused course and minor.
Announced two weeks ago, the AI Makerspace supercomputer will initially be used by undergraduate engineering students at Georgia Tech. Ultimately, however, the hope is that access to computing resources, typically prioritized for research, will be democratized across all universities.
computer world I talked to matthew blockThe associate dean of Georgia Tech's College of Engineering talks about how new AI supercomputers will be used to train a new generation of AI professionals.
Below are excerpts from that interview.
Please tell us about the Makerspace project and how it came about.? “Makerspace was actually the vision of Arijit Raychoudhury, our dean and chair of electrical and computer engineering (ECE), who really wanted to put AI into the hands of his students.
“In 2024, in a post-ChatGPT world, things are very different than they were in a pre-ChatGPT world. Doing anything meaningful and relevant to the industry requires a lot of computing power. And in some ways, the devil… is out of the box. People understand what AI can do. But I think you need infrastructure to get to that level of training.
“The name Makerspace also comes from the culture of makerspaces at Georgia Tech.Makerspaces are places where students can tinker inside and outside of the classroom. The idea is to give them the tools they need to do AI in a way. So we're currently partnering with Nvidia to basically give students a supercomputer. is.
“What's unique about this is that it's aimed at supporting students. And now they're in the classroom. We're still rolling out. We're in phase one, which means students in the classroom are working to support the industry. The idea is that you can work on AI projects, problems that are interesting from a pedagogical perspective but don't make much sense in an industry setting.”
Nvidia H100 Tensor Core GPU in Makerspace
Georgia Institute of Technology College of Engineering
Can you tell us a little bit about the project they're working on this?. “Let me give you a very specific example. ChatGPT is a very typical and very specific form of AI called generative AI. As you know, it can generate. [that means] Text that responds to the prompt. You may have seen generative models that generate images. I think these were very popular. These are the kinds of things our students can do right now, producing something photorealistic, for example.
“It takes a lot of computing power to train a model and test that it's working properly. That's what our students can do. We've seen how much progress we've made. To find out, before AI Makerspace, students primarily relied on something called Google CoLab, which gave them free access to some computing resources. They're actually giving us resources that they're not using or selling to their customers.
“They're very nice.” [Google] There are only so many things you can do to do that [limited resources]For example, if you want to train on around 12,000 images. For example, you can now train generative models on datasets containing as many as 1 million images. So it can actually scale up by orders of magnitude. You can then start generating photorealistic images that were previously impossible to generate. This is the most visual example I can give. ”
Can you tell us a little bit about the genAI project that your students are working on? How good is the technology that will produce the results they want? “That's a hard question to answer. I mean, there's a lot of layers. We just launched, literally like AI Makerspace officially opened two weeks ago. So right now… It's actually being used on a large scale in the classroom, where students are learning how to do machine learning. [The students] I need to get the data. [They] You have to learn how to train the model. The students have a homework project that consists of this fairly sophisticated model that they need to train and test.
“Now we have a vision beyond that, what we're calling Phase 2 of the Makerspace. We're doubling the computing power. What we're thinking now is turning it into a senior design project. We're going to expand that into what we call vertically integrated projects, where our students basically do multi-year long-term research. is going to do many things – all of which of course [the] engineering [school].
“We've encouraged a lot of faculty to create a lot of new courses across engineering schools in AI and ML that are important to their fields. For example, if you're an electrical engineer, there's a lot of hard work involved. You know you have hardware and you know you have a model for it. How do you make the model small enough to fit into the hardware? That's a very specific question that students will have. , for example, a mechanical engineer might use it differently. Perhaps for them, what generative AI can do is generate 3D models and help them think about structures that they wouldn't think about naturally. And you can reject that model. However, how the tool is used really depends on the capabilities of the particular makerspace. to be available beyond engineering.
“It's already being used in our School of Computing, and we hope that our colleagues in the School of Business and elsewhere will see the value in it, because they haven't used AI yet. For example, financial models and predicting whether to use “sell or buy stocks.'' I think the sky is the limit. No one uses AI through Makerspaces. An infrastructure that provides tools. And these tools find biases in all different areas of expertise. ”
Why is it important to implement this technology in schools to help students learn about AI?Here's how we articulate this: We are not the deliverers of an apocalypse scenario where AI creates terminators that wipe out humanity. Well, that's not what we think.
“AI is definitely going to change things. And we think it will definitely displace a minority of people. AI-enhanced humans will replace non-AI humans. think.
“I think a lot of discussion has formed since ChatGPT was released to the world, and sometimes there's great fear in universities. Are students cheating on essays? Are students cheating on this? I had this discussion with a colleague in the computing field, where we have an introduction to computing class where we cheat and write code, but this is not the correct approach. I don't think so. But the devil is out of the box. It's a tool here and we have to learn how to use it.
“To use my own analogy, I'm driving a car. I don't know how my car actually works. I mean, I've never been a mechanical or electrical engineer. I kind of know if it's necessary [for a car to run], but you can't fix it. But that doesn't mean you can't drive. And I think we're at that point with AI tools. You don't want to be the one riding a bike when someone else has a car, so you need to know how to use it.
“Not everyone needs to be a mechanic, but everyone needs a car. So we want every student at Georgia Tech to know how to use AI. What that means depends on your field and major, but these are tools and you have to play around with them to really use them.”
How has AI enhanced Georgia Tech's curriculum? “In a way, we were lucky.” [we’re] We will build that infrastructure from scratch. But when you think about AI, Georgia Tech has been working on it for decades. Our faculty has a strong focus on research. They are doing cutting edge research, and AI has always been in the background, at the roots of AI. We had a lot of colleagues who were actually doing machine learning, even if they didn't say it in these terms.
“Then when deep learning started coming out, people were ready to understand it. So we were already thinking about doing it in the lab, and the integration into the curriculum was already slow. So what we decided to do was accelerate that and create a makerspace that was needed to incentivize teachers and reimagine their curriculum with AI and ML in mind. It was about accelerating other mechanisms.”
So what kind of AI courses have you launched? “I can give you two examples that we've launched, and they're very new. But I think we've already been pretty successful. One is that we've officially launched an AI miner. That's what I did.
“The great thing about this AI miner is that [is that it] This is a way for students to take a series of courses with a consistent, unified team, with credits listed on their diplomas and transcripts. This minor is currently designed as a collaboration between the College of Engineering and the College of Arts and Sciences.
“Then there's the ethics and policy part. Students have to take a course specifically designed on AI ethics and AI policy. We think very holistically. Just for fun, if we just train engineers to do the technology part on their own, we're likely to see Doomsday and Terminator scenarios happen.
“We want our students to think about the uses of AI because it is a technology that can be used in many ways. [and problems associated with it]. Let's talk about deepfakes. We worry about it for all sorts of political reasons.
“Another thing we've been doing in the College of Engineering is basically encouraging faculty to create new undergraduate courses that are related to AI and ML, but also related to their respective disciplines. is literally [just made the announcement] The university approved 10 new or significantly improved courses. What this means is that we have courses in machine learning for smart cities, courses in civil and environmental engineering, and chemical processes in chemical engineering and bioengineering, where we are using AI and ML for completely different purposes. That's how we think about AI. It's a tool. So your courses should incorporate that tool. ”
Are your students already using genAI to help them create applications, i.e. software engineering and development? “Formal or unofficial? There's no good answer because we don't really know. But what I do know is that students use it whether they're with us or not.” So you know they're using generative AI because they all have subscriptions to ChatGPT.
“In the context of a Makerspace, this is a resource that can start all sorts of things. Our students use it to write lines of code.”
So what do you think are the most popular uses for AI Makerspaces right now?“It hasn’t been long since we officially started this on a large scale, so I can’t prove it. This is mostly in a classroom setting, where students are learning things they wouldn’t have dreamed of before.” Used for different types of homework.
“We plan to launch this and use it over the summer for an entrepreneurship program called Create So this is going to be used primarily over the summer, and we've partnered with Nvidia to test it over the last few weeks in the context of a hackathon where teams have big problems they want to solve. And we want to accelerate their science by teaching them how to use that makerspace, in NVIDIA's words.”
