About Me
I am a PhD student supervised by Roger Grosse at the University of Toronto. Very broadly, my research goal is to help guide the development of artificial intelligence in a way that benefits the greater good. I believe AI alignment is necessary (but not sufficient) for this, specifically alignment towards a broad set of values representative of as many people as possible. I have been working towards this goal in baby steps.
Research
My most recent paper is on applying twisted Sequential Monte Carlo for controlled generation from language models (won a Best Paper Award at ICML 2024). In addition to providing alternative learning methods which may outperform standard methods such as PPO (whether this will scale to larger models is a question for future work), this gives us a richer quantitative evaluation of language model outputs via KL divergences from the desired output distribution. The hope is this work provides new methods for aligning language models or revealing undesirable behavior, and under certain (possibly restrictive) conditions, evaluating how aligned our model is.
I completed my MSc in September 2022 at the University of Toronto, supervised by Roger Grosse and Jakob Foerster, working on more consistently learning reciprocity-based cooperation in social dilemmas. This work is hopefully a step towards ensuring interactions between AI agents result in socially good outcomes.
For all my publications, see my Google Scholar.
My Journey
I did my first undergrad (2011-2015) in Finance and Economics, as I found economics most interesting among my high school courses (which notably, did not include computer science). As I did finance internships, I developed an interest in programming for automating repetitive work. This eventually led to an interest in computer science, and I took my first CS course in my fourth year of study. As interesting as it was, I couldn't bring myself at the time to change careers and give up my years of study in finance as well as my prestigious finance job offer. I found my finance work interesting at first, but then began to question whether this was what I wanted for the next 20-30 years of my life. This led me to exploring my passions in my time outside work; I tried competitive gaming (reaching top 60 in North America in Hearthstone in Feb. 2016), but found treating gaming as a job ruined its fun. I then tried game development, my childhood dream, which culminated in the release of Path To Oblivion after 2 years. The experience wasn't bad - I would consider game development again in the future if I had time - but during that time, the developments in AI (starting with AlphaGo, as I was a Go player) caught my attention. Eventually, after having taken night courses to get my feet wet, I decided that AI research was the perfect intersection of my interests, strengths, and potential societal impact. I quit my finance job in 2018 to return to undergrad studies, this time in computer science, with the goal of conducting AI (alignment) research. I've been privileged to have the circumstances that allowed me to make the decisions I did, and very fortunate to have had the mentorship, guidance, and opportunities that have led me to where I am now.
While I wish I started exploring computer science earlier, I learned a lot from my previous life in finance as well - about corporate life, business, interpersonal relations, communication, leadership, teamwork, and most importantly, myself. When doing research work or studying, it's easy to get lost in the weeds; I try to remind myself to periodically step back and look at the bigger picture, reminding myself why I'm here, doing what I'm doing. I'm here for myself, and for everyone.