Dr. Peter West joins UBC Computer Science to lead research program in large language models
New Assistant Professor Peter West, Ph.D. to study natural language processing and generative AI systems
When Peter West entered his first year of undergraduate at UBC over a decade ago, he had no plans to study computer science.
“I was originally in physics,” he says. But on orientation day before the start of classes, someone told his cohort to take the introductory computer science course to know a bit of programming. “That course was what got me into coding — it was very intuitive to me.”
By the end of his degree, he was an Honours Computer Science major, eager to apply to graduate school to continue his education. While he completed several stints of research in optical physics and computational biology, he gravitated most towards the intersection of artificial intelligence and language.
“I’ve always been interested in communication and language — how people communicate with each other and the ambiguity of it. It’s probably from reading a lot and talking to a lot of people as a kid,” he says.
He went on to do his Ph.D. at the University of Washington, starting off by studying how to generate training data from large language models and building new algorithms for smaller language models.
But a few years into his degree, the boom in these generative AI models upended his research plans.
“All of a sudden, these language models could solve a lot of tasks that we, as researchers, were interested in,” Dr. West says. “My research near the end of my Ph.D. became more about understanding the limits of these models.”
Knowing how well these models can perform at some tasks but not others is important when deciding what to use them for, according to Dr. West.
“If these models are effective 90% of the time but make odd mistakes that we don’t understand 10% of the time, people will probably end up trusting those mistakes,” he says. “The goal is to help people develop a better intuition of where these models break down. There’s a natural tendency for people to completely trust them, even in places where they’ve never verified that it gets the right answer, such as giving life or financial advice.”
He continued studying the limits of language models during his postdoctoral fellowship at Stanford University and he plans to carry these research plans forward as an Assistant Professor at UBC’s Department of Computer Science. In particular, he hopes to understand and evaluate language models and find new sets of evaluations that cover “edge cases,” or situations in which the models encounter a problem when asked to perform a unique task.
“It’s very exciting to be able to continue working on these scientific problems, even if they may not be as relevant at first,” he says. “That’s how we got things like language models and neural networks — because people in academia had an idea and were willing to put decades of effort into it.”
Dr. West’s decision to return to UBC was an easy one. In addition to his love of Vancouver, he’s excited to be back at the Department of Computer Science, now as a professor.
“The department has friendly and supportive people who are also ambitious and are leaders in their fields,” he says. “I love working with academics — they’re curious people so you have interesting conversations every day.”