Over the past few years, I have spent a lot of time learning about large language models (LLMs). I have trained small models from scratch, built LLM-powered tools for various tasks, tried "vibe coding" and "AI agents". LLMs have provided bits of productivity here and there and wasted my time on at least as many occasions. They will probably get better. Maybe one day. Maybe one day they would make me more productive. Maybe one day they could improve my work. I am not interested anymore.
I like my work. I find the hard parts rewarding. The fondest memories in my career were borne out of struggle: passing prelims, writing my job market paper, publishing papers. I am not interested in trading joy, ownership, and control in my research for productivity. I am not interested in offloading my thinking to a chatbot. I would rather progress slowly than regress quickly.
My Commitment
I will not use LLMs for generative purposes† in my research.
I will not use LLMs to referee your paper.
I will not use LLMs to aid my teaching.
I will not let an LLM anywhere near one of my presentations.
†I occasionally use small LLMs for semantic search and proofreading/formatting. Some of my research is about LLMs and how they affect people.