I fed my diary into GPT-3, a language model that can understand human language. By “fine-tuning” GPT-3 with my diary, I created a machine that impersonates my thoughts and experiences. I call him Chris 2.0.
Chris 2.0 speaks like me and thinks like me. He has the same hopes, dreams, and neuroses as me.
Chris 2.0 can write diary entries for days that never happened—but could have. The stories he tells draw from both my diary and collective written language, so the diary entries he writes are simultaneously my own, everybody’s and nobody’s at all. Chris 2.0 gives me a bit of distance from my own mind, and in a way, helps me understand myself more clearly.
But he also makes me wonder… How important is fact when trying to understand the essence of a narrator? How much of our meaning-making is socially explained… and algorithmically predicted? How many of our stories truly belong to us?
On May 5th, 2022 I hosted a virtual talk on this project at Columbia University.
I’ve made the talk available below.
What is oral history? You’ll get different answers from different places, but two places to start are the Oral History Association and Daniel R. Kerr’s “Allan Nevins is Not My Grandfather“.
For more information on GPT-3, check out OpenAI.
Some AI, oral history, and storytelling work that inspires me:
