Amanda Butler ’18 works at one of the most consequential addresses in technology. As a Higher Education and Research Solutions Architect at NVIDIA, she helps university researchers across the country—computational linguists, anesthesiologists, chemists, astrophysicists—take advantage of what the company calls accelerated computing: the ability to run complex tasks exponentially faster than traditional processors allow. Most people know NVIDIA primarily for its GPUs, the specialized chips that have become the engine of the AI boom, but this is only one piece of the puzzle. Accelerated computing touches everything from video game graphics rendering to molecular dynamics simulation. Think of it, as Amanda does, like a layer cake: at the bottom, the hardware—chips, GPUs, interconnects. Built on top of that, drivers, operating systems, code libraries. On top of that, applications. The goal in creating every one of these layers is the same: make sure researchers and developers don’t have to reinvent the wheel every time they want to solve a hard problem. Amanda’s job is to make sure the scientists who need to tap into these technologies can do so effectively, with a line straight to the source. 

On any given day she might be giving a lecture on emerging hardware optimization tools to a research computing team, sitting side-by-side with a physicist to find bottlenecks in their code, or working with a university’s IT department to build the guardrails that keep an AI system on task and secure. It is a role that defies easy categorization—part engineer, part educator, part translator between the world of industry and the world of the academy. She describes herself, with some amusement, as “a subject-greedy nerd” who could never quite commit to a single domain.

One of the first things Amanda wants to clear up is that AI is not synonymous with chatbots. That’s one application—a visible and accessible one—but underneath it is a far broader technological shift. In her work with universities, Amanda sees AI being deployed in ways that would surprise most people: reconstructing the text on ancient scrolls without having to unroll and potentially destroy them; modeling novel protein structures; running many-variable climate simulations across vast timescales. These often aren’t chatbots answering questions in plain English—they are systems learning the underlying structure of a physics problem or a crystal lattice, finding patterns in data that human researchers might take decades to detect on their own.

And when a university does want to utilize large language models, managing all of this responsibly is its own significant challenge. When Butler works with university IT teams on deploying AI tools—whether a help-desk chatbot or an autonomous research agent running in the background—a major part of the conversation is about guardrails: how do you keep a system on task? How do you ensure it’s drawing only from approved sources rather than inventing answers or pulling from the open web? One increasingly common solution is the “sandbox”: a carefully defined environment that limits which applications an AI can access, so researchers can let a system run autonomously without worrying it will wander somewhere it shouldn’t. These are the unglamorous, essential problems that sit behind every impressive AI demo.

Amanda didn’t start coding until she got to Yale. At College Prep, her formative experience was freshman physics with Dr. Mike Lane—Doc, as students called him—whose background in astronomy quietly reshaped her sense of what was possible. Radioactive decay, particle interactions, a field trip to Griffith Observatory and Joshua Tree: by senior year she was taking both an advanced physics course and an astronomy elective, and her trajectory was set. The technical skills came later, in college and beyond, when she moved from Python to C and C++, eventually finding her way into high-performance computing through both coursework and research.

That sequence—curiosity first, technical fluency second—turns out to be a reasonable model for navigating a field that changes as fast as this one does. Even people who have spent decades in tech can find themselves starting from scratch when a new technology emerges. What carries you through isn’t any particular skill set; it’s the confidence that you know how to learn. College Prep, she says, gave her that.

 

Which brings her to the question students are actually asking: given all of this, what’s the point of doing the hard thing yourself?

Amanda is honest about the pull of convenience. AI makes it genuinely easier to produce something—a draft, a summary, a block of code—without going through the full process of producing it yourself. But she argues that the struggle is where the learning lives. The late-night essay and the grinding problem set are, at their core, teaching the same thing: how an argument is constructed, how to show your work, how to tell the difference between a claim that holds and one that doesn’t. Those capacities matter more, not less, in a world that is generating more content, more opinions, and more plausible-sounding answers than ever before.

She’s also direct about failure. AI makes it easier to avoid it, or at least to fail privately. But the willingness to try something genuinely hard, to sit with not knowing, and to work through it anyway is a skill—one that atrophies without practice, and one that will still matter when the AI can’t get you the rest of the way.

Her preferred frame for students isn’t necessarily avoidance but intention. If a student chooses to use AI, she posits, it should be as a sparring partner—something to argue against, to pressure-test your thinking, to find the holes in your own argument. Not to produce the argument for you, but to make yours better. And for students in any discipline, not just STEM, she recommends developing at least a basic understanding of what AI actually is under the hood: that it is, at its foundation, data science, statistics, and computer science. Not because everyone needs to build their own model, but because understanding the mechanics helps you understand the limits—and that’s the difference between being a thoughtful user of a powerful tool and being someone the tool is using.

The tools will keep changing—faster than any curriculum can track, faster than anyone can fully prepare for. What doesn’t change is the need for people who can think carefully, argue rigorously, collaborate across disciplines, and adapt without panic. A rigorous education isn’t in tension with that future. It’s the preparation for it.