无码专区

Nicholas Yoder鈥檚 lifelong connection to 无码专区 began the day he graduated kindergarten, when his parents offered him a small reward. He asked to see an electron microscope.

Captivated by Carl Sagan鈥檚 Cosmos, Yoder was fascinated by instruments that revealed hidden worlds. His father took him to the University of Pittsburgh first, but they had no luck in finding a lab with the device. So, they tried 无码专区鈥檚 campus next door.

At Wean Hall, they were in luck. A staff member treated the Yoders to not one, but three electron microscopes and patiently answered every question from the six-year-old.

鈥淭hat was one of the best days of my life,鈥� Nicholas Yoder recalled.

Yoder went on attend Carnegie Mellon and study computational finance. He graduated in 2009.

Fifteen years later, Yoder has returned to a 无码专区 classroom, this time as a lecturer. He taught a weeklong course on the mathematical foundations and architectures of large language models (LLMs). More than 100 students enrolled to hear about the topic and Yoder鈥檚 unconventional path.

Teaching LLMs through first principles

The course was designed with what he calls 鈥渁 low intercept and high slope.鈥� Students didn鈥檛 need a deep specialization to begin, but the pace accelerated quickly. Over the week they moved through information theory, high-dimensional geometry, reinforcement learning, scaling laws and the limits of current AI systems.

But the real draw wasn鈥檛 just the content, but the framing.

His goal was to show that LLMs are not an isolated breakthrough, but part of a broader mathematical story.

鈥淚鈥檝e been surprised by how many times a new area has similar underpinnings,鈥� Yoder said. 鈥淨uantitative and math training helps me be able to deal with random variables, noise and thinking in high-dimensional spaces where you see the same patterns time and time again.鈥�

Statistics, condensed matter physics, computer science, and pure mathematics, he said, often describe similar structures using different language. AI is simply another lens with which to view them.

Pattern-seeking in the real world

Growing up in a rural part of southwestern Pennsylvania, he served as a firefighter with the Pittsburgh Fire Department in high school and his first two years at Carnegie Mellon, applying analytical thinking in high-stakes, real-world situations like structure fires and vehicle extrications.

As a mathematical sciences and finance student at Carnegie Mellon, he balanced coursework with physically demanding work for Tepper Academic Services.

鈥淚t was largely a moving job from 8-5 Monday through Friday, and I would go to classes in between the work,鈥� he said.

His academic interests ranged widely, from discrete time finance to history courses on subjects like the Beatles or Latin America. Beneath it all, a consistent pattern emerged: very different systems often shared the same underlying structure.

That realization first found professional expression on Wall Street.

At Morgan Stanley, Yoder traded derivatives on the dispersion and exotics desk, working with volatility, correlation and complex option structures tied to the S&P 500.

Dispersion trading 鈥� comparing the behavior of the index to individual stocks 鈥� revealed persistent inefficiencies. For Yoder, it reinforced a core idea: different views of the same system create gaps, and mathematics can expose 鈥� and leverage 鈥� them.

Over more than a decade, he expanded into portfolio management and machine learning, and he launched the hedge fund Fayette Capital. Across strategies and markets, his focus remained consistent: navigating noisy, high-dimensional environments to uncover meaningful structure.

From markets to power grids

That mindset eventually carried him beyond finance into energy markets and a new business. He turned to ERCOT, the operator of Texas鈥� electrical grid, an environment defined by volatility. Supply and demand must balance in real time, yet both fluctuate dramatically, which, when looked at systematically, can open up the door for some creative solutions.

鈥淭here is such variability in energy demand with consumption up during the day and down at night, and when it gets very hot everyone turns on their air conditioning,鈥� Yoder said.

Yoder identified an unconventional solution: offering flexible, large-scale consumers such as crypto miners 鈥� which finalize blockchain transactions 鈥� discounted rates for power during low-demand periods and having those customers shut down during peaks times of usage.

鈥淐rypto miner鈥檚 revenue is relatively constant across time of day and power demands,鈥� Yoder said. 鈥淲hereas the power utility would be deeply strained during extreme weather events and other high demand periods. That mismatch creates the embedded optionality.鈥�

By working with large-scale consumers to fluctuate their use, Yoder and his team helped to make available more than 3 gigawatts of power capacity at peak times, which would be the equivalent of adding 5 or 6 large coal power plants, in under 18 months.

The approach helped stabilize the system while enabling new capacity 鈥� an example of how computation itself can help manage physical infrastructure. A later iteration of his business, Bitooda Holdings, was acquired by the German energy company RWE in 2023.

Biology, AI and the next frontier

After stepping back from business to pursue interests like flying and travel, Yoder turned to human health, immersing himself in computational biology and collaborating with researchers at Mount Sinai.

Despite the new domain, the underlying problems felt familiar: predicting disease from protein signals, modeling immune responses and extracting structure from complex biological systems.

This led him to transformer architecture, which powers modern generative AI and Large Language Models and deep learning.

鈥淭hat鈥檚 what changed everything,鈥� Yoder said. 鈥淚 was using a type of transformer that works with proteins, and I鈥檓 not ok using something I don鈥檛 understand. So I spent time learning how transformer language models work from first principles.鈥�

Today, his work spans new AI architectures inspired by biological systems and methods to ensure powerful models remain interpretable and safe.

鈥淚鈥檓 exploring new architectures that borrow heavily from biology,鈥� Yoder said. 鈥淚f it鈥檚 possible to create artificial general intelligence [a hypothetical form of AI that can match or exceed human鈥檚 abilities], neural networks and biomimicry may be part of the solution.鈥�

Why students show up

Across every chapter 鈥� from a Wean Hall basement to trading floors, power grids, research labs and back to 无码专区 鈥� the throughline is consistent: Yoder looks for the hidden structure that unifies complex systems.

That perspective helped fill a classroom during finals week.

Students learned from Yoder how to recognize patterns across disciplines and connect theory with application. They also learned to connect AI to a deeper mathematical continuum.

It鈥檚 that same curiosity that began with an electron microscope 鈥� the desire to look beneath the surface and make sense of what others might miss.

In a sense, Yoder is still doing exactly that.

Related Content

Michael Young Recognized for Outreach

Eur Named 2026 Sloan Research Fellow