My fast-evolving perspective on AI in academia

A few minutes ago I made a two-line change to my advice to students that ask me for a letter of recommendation. It’s a pretty minor change but it reflect how my perspective on the use of AI in academia is changing very rapidly. In 2024 (the ancient prehistory in AI land), AI-generated text was inferior to a well-crafted human letter and reading this kind of slop was a quick way to a rejection. Nowadays, Gemini will take in a couple of example letters, some student particulars, and will quickly generate a near-perfect letter, reproducing my writing quirks in the process.

Why then not leave the writing process entirely to Gemini?

I am not blind for the incredible advantages that AI offers: at this moment, 5 Claude terminals are busy combing my lecture notes for mistakes, writing and running my research code, and one is even working through my credit card bills on the side.

At the same time, my role as a professor is quickly changing: I want to equip students with a similar kind of (excellent) educational foundation that I received

), both to myself and more so towards the students. At the same time, my role is radically changing:

Text is king

Low cognition tasks are dead