Ariana Thornton

Student Interview: John Guerrerio '26 on AI Research at Dartmouth

John Guerrerio '26 didn't always want to work with AI. It was at Dartmouth where he fell in love.

Hailing from Baltimore, Maryland, John originally wanted to pursue cybersecurity. His interests shifted during the AI boom of the early 2020s, leading him to pursue award-winning research in AI safety at Dartmouth.

When I sat down with John in Baker-Berry Library, he had just come out of an intense week finishing a paper developing a methodology to improve Large Language Model (LLM) in-context reasoning to submit to ICML, a leading machine learning conference. He enthusiastically recounted his journey in AI and his professors' positive effect on his growth.

This interview is edited for clarity and length.

Ariana: What brought you to Dartmouth?

John: I visited a lot of colleges when I was applying. What stuck out to me about Dartmouth is that in class, you actually get to interact with professors in a one-on-one setting. They know your name and who you are. If you want to do research, it's really nice to have that relationship.

Ariana: Where do you think Dartmouth sits in the world of AI?

John: We're the university that invented the term artificial intelligence. We've been expanding the Computer Science Department in the past couple of years, especially in AI, in ways that I feel are really beneficial. [Note: As of January 2026, Dartmouth undergraduates pursuing the Bachelor of Engineering (BE) can now choose a concentration in AI.]

The 1956 Dartmouth summer conference brought together visionary thinkers in computer science, mathematics, and cognitive psychology to launch the field of AI. (Source: IEEE.org)

Ariana: You applied to Dartmouth thinking you would go into cybersecurity. Was there a defining moment when you realized you wanted to pivot to AI, or was it gradual over time?

John: Definitely gradual over time. It was less of a moment and more of an experience.

One of the pivotal experiences in this transition was when I joined Professor Soroush Vosoughi's lab. He's a computer science professor here who works on pretty much everything related to natural language processing. I was working closely with him and one of his PhD students, Weicheng Ma, who is now a professor at another university.

The project I joined studied bias in Large Language Models (LLMs). In systems security, you're trying to figure out where the bug is and how to fix it. LLM bias is the same idea—you have some hidden flaw that's causing harm, but it's not clear how to measure or eliminate it. It was all the problems I really liked about cybersecurity, but in a setting that I found more interesting. 

Ariana: How did you get involved as a data developer at Dartmouth's Digital Applied Learning and Innovation Lab (DALI Lab)?

John: I had done a lot of AI research and wanted to try something more applied, so I applied to DALI Lab.

During my first term there, I worked with the National Park Service on models that could automatically count barnacles in pictures of tide pool samples. That had been a really tedious task they had to do by hand—there's hundreds of barnacles in these images—so we reduced the time they needed to spend on this task.

In my second term, I worked with the Dartmouth-Hitchcock Medical Center on a project called Smart Microscope. Essentially, they were trying to create a microscope that could use digital pathology. My project integrated AI features into the microscope and created a model that could simulate certain types of staining, which could be used to diagnose cancer.

Students at DALI Lab work in small teams of engineers, designers, and project managers.

Ariana: Tell me about the research you did that won the Barry Goldwater Scholarship!

John: That was the LLM bias research I did with Professor Vosoughi.

When an LLM is trained on data scraped from the entire Internet, it inevitably learns to replicate very toxic ideas. In order to unlearn those ideas, you first have to figure out what stereotypes it has learned. Most methods to do that are very data-driven, so if you only use data from an American cultural context, you miss stereotypes in other languages and lower-resource cultures.

Our work contributed, to the best of our knowledge, the first native Spanish language stereotype examination dataset specific to four countries in Latin America and Spain. We used it to benchmark two language models for bias. I think our most interesting finding is that biases for different cultures are encoded in different parts of the models we looked at. 

Ariana: How do you hope to make an impact in AI in the future?

John: I'm currently applying to PhD programs to grow as a researcher in the field. I ultimately hope to become an academic with some ties to industry.

I think Professor Adam Breuer is a good model of how I hope to structure my future career. He's a professor here, but his lab is funded via OpenAI and his findings are applied by the company into models like ChatGPT, which is an algorithm that billions of people use. I hope to be doing research in a way that allows me to make the models that everybody uses safer.

Ariana: What advice would you give to your past self and to students hoping to get involved in AI at Dartmouth?

John: The advice I would give my past self is to take time to explore. Back then, I erroneously believed that I needed to specialize as soon as possible to be competitive for graduate school. Try lots of different areas to figure out what you like and don't like.

The advice I would give to students who are looking to get involved in AI research here is that the professors here can seem intimidating—you look on Google Scholar and some of them have 15,000 citations—but they're all really nice people. Take a class with them, get to know them, and build a relationship with them. That can turn into a project, which can turn into an ongoing collaboration.

Ariana: Do you have any concluding thoughts you'd like to share?

John: It's really worth stressing how fantastic the mentorship is here. That's why I came to Dartmouth, and that's honestly one of the reasons I'm still in this field. I was very fortunate to have such fantastic mentors like Professor Soroush Vosoughi, Professor Yaoqing Yang, and Professor Adam Breuer, who took a chance on me to do research with them and have been so supportive in navigating PhD applications.

For instance, my first PhD interview was scheduled for the day after Christmas. I got the invite the week before and messaged Professor Yang to ask how it worked because I had never done such an interview before. He responded with such fantastic advice even though it was over the holidays. I'm really grateful to have mentors like that who are so incredibly supportive.