Romeo Dean is a senior studying a concurrent bachelor’s and master’s degree in Computer Science. He is a part of Harvard’s AI Safety Student Team (AISST). This interview was conducted during the Fall 2024 semester. 

HTR: What does being an innovator mean to you, and how does this manifest in your daily life?

Romeo Dean: Being an innovator for me means a combination of having ambition, but also humility and curiosity. So the ambition to look at things and be critical even if they’re big things in the world, and asking why are they like this? Why aren’t they better? Why couldn’t they be this other way? But then, at the same time being humble and curious as to the things that you realistically do know and don’t know, and having reverence for the expertise and things that are established and known in the world, and trying to understand them. So I think it’s a balance of these two things. 

In my day to day life, which has now become so focused in this area that I’m into — AI safety — it’s about keeping that ambition to make a large impact, but also having the determination to be learning from all the experts and knowledge that’s around and available at Harvard and beyond.

HTR: Along the same lines, what sparked your interest specifically for AI safety, since that’s what you’re specializing in?

Romeo: I came into Harvard, and I had various interests. I think it even started with more economic things, and then mathematics — and I really enjoyed my computer science classes. But AI, in particular, was always very fascinating to me from a distance, for the little I knew about it. I kind of didn’t have a direction or a clear sense of what I wanted to do — I had a vague idea of wanting to try and use AI in some way. 

But then, what really sparked my interest was getting involved with the student group here on campus. The AI Safety Student Team has an intro Fellowship, which really impacted me. I became convinced that “hang on before, I can think about what I personally want to do, there’s actually a serious chance that things get messed up.” There are incentives at play to develop AI quickly and there are also a lot of unknowns. I think there’s potential for there to be unchecked and uncontrolled power with this technology. And I think this really motivated me to to have this focus of “how can I do my part to make sure that we navigate this phase where we’re still developing AI” — and the more I learned, the more I came convinced that now is a really crucial time, and things are moving very quickly, much more quickly than it than it seems on the surface. And that’s kind of in the branch that I have gone down.

HTR: Yeah, that’s really cool that the student group is aware. Because I think, especially nowadays, a lot of people idolize, maybe even abuse, AI. So the fact that you guys are aware of and know its consequences is really cool. 

You mentioned the student group, but do you think any classes at Harvard sort of sparked this interest?

Romeo: Out of my classes at Harvard, I think I just really enjoyed the format and the organization of the classes in the computer science department. And the more I took computer science and other classes at the same time, it just became more and more clear to me that computer science was the thing that interested me. 

I think, in particular, I really enjoyed learning this semester, actually, about system security. So it’s a class with Professor James Mickens that basically covers all the different kinds of security vulnerabilities across the internet, across everything from satellites to cryptocurrencies. This is now increasingly tied to AI in a really interesting way. It might be some recency bias, but I think it’s by far the class that I’ve both enjoyed and learned a lot from Harvard so far. 

HTR: Would you see Professor Mickens as a mentor? Or is there anyone on campus that you see as a mentor? 

Romeo: Yeah, I am hoping now that I’ve taken this class with Professor Mickens, that that will develop. 

But I think if I had to say a mentor, it’s actually been from the start of last year, when I joined a mentorship program that I found out about through AISST. But it was run by an organization in Berkeley, California, and it matched mentors in AI safety with students. I’ve been matched and have been working with Daniel Kokotajlo since. He was recently in the Times 100 for Most Influential People in AI. When I was matched with him, he was still at OpenAI, but he’s since left and started his own organization because he became concerned about how he thought OpenAI would be navigating the next few years. The new organization is called the AI Futures Project, and I’m starting to officially work with them part-time right after the semester ends.

HTR: Is that somewhere you’d like to work after graduation? 

Romeo: Yeah, I think so. My plan is to be working part-time through next semester — my last semester — now that I’ve finished most of my class requirements. But yeah, after I graduate, I think I would really enjoy working there — I think I will continue to find the work that Daniel and his organization plan to do, really exciting and important.

HTR: That’s exciting, good for you! Diverting a little bit — would you mind sharing a key story, challenge, or success that you think has contributed to your innovation? 

Romeo: I think probably the story that feels most important to me has been the last year and a half. It was when I was at Harvard and I knew the classes I liked, but I didn’t know sort of what I wanted to do with them and how to apply myself; I went from basically having no idea about what AI safety was to very quickly becoming a part of AISST — and then joining this mentorship program with Daniel. 

From there, it seemed to accumulate, and have this snowball effect where I’ve been forecasting and trying and learning about all these trends in both the hardware and software side of AI development. Some of my models of how compute production is going to go over the next few years have gotten wide attention in the community, and I’ve been invited to talk about my modeling at RAND, for example, which is a think tank in DC. I’ve held sessions at two different AI security forums about different kinds of tabletop exercise scenarios and how AI development could go. I’ve also run some AI capability benchmark forecast workshops at AISST. And then I helped with the recent launching and question design for an AI 2025 forecasting survey. 

And so I guess, it’s been this really, really intense period from basically knowing nothing about this field, to just being intensely bought into it and carving out the niches that I’ve been happening across, and being interested in and and feeling like I’m able to have a lot of impact in the space. 

HTR: It sounds like your work is becoming increasingly influential —  clearly, your classmates think that you’re a top 10 innovator! I have one last question for you: What advice would you give your younger self about navigating Harvard?

Romeo: That’s a good question. I think one really important thing about navigating Harvard is that Harvard has so many different opportunities and with those opportunities, I think, there’s a natural form of pressure. It’s really easy to confuse a healthy form of exploration with a kind of unhealthy pressure to compete or succeed. I think it’s a difficult sort of idea — and there are real reasons to not do this — but just the idea of always trying to choose what you do and don’t engage with, and the classes you do and don’t take, and the activities you do and don’t join, based on what you are actually interested in, think is important, and think is going to fulfill your goals, aspirations, and also your values. So yeah, I think at times I would have reassured my younger self not to give into certain pressures or to follow paths that seemed better from a success or a status perspective.

HTR: Thank you so much for taking the time to interview with us!