Faculty at Home

• Phil Chodrow

Large language models (LLMs) power modern chatbots and text generators. How do they work and what do they mean for human society? We’ll begin by describing next-token-prediction and reinforcement learning with human feedback, the two technical tasks that underlie contemporary LLMs. We’ll then turn a critical eye toward claims about the capabilities of LLMs, especially those propagated by large tech corporations. We’ll close with a discussion of the impacts of LLMs on labor, the environment, artistic production, and cultural representation.

Phil Chodrow is an assistant professor of computer science at Middlebury College. His interests include network science, models of social systems, data science, and critical computing pedagogy. He received his PhD from MIT and his BA from Swarthmore College.

• Megan Mayhew Bergman

In Vladimir Nabokov’s memoir of estrangement—Speak, Memory—he writes: “I see again my schoolroom in Vyra, the blue roses of the wallpaper, the open window… . Everything is as it should be, nothing will ever change, nobody will ever die.” This talk examines the flow between person and place—how we are shaped by the landscapes we inhabit, and how, in turn, they shape us, physically, emotionally, and creatively. We will discuss the impact that loss and environmental degradation can have on one’s sense of place and belonging, and how that change might register in memory and creative output.

Megan Mayhew Bergman is an assistant professor in creative writing and director of the Bread Loaf Environmental Writers’ Conference at Middlebury. She is an award-winning fiction writer and climate journalist who has been published in the New Yorker, the Atlantic, and the New York Times, and has published three books with Scribner. She recently founded Open Field, a nonprofit geared toward increasing the accessibility of environmental storytelling and advocacy skills, and is the cofounder of GreenStory, an environmental narrative agency.