While I’m the first author, the students were truly incredible. They were instrumental in sanity-checking the tasks and surfacing edge cases. They asked me questions I had neglected to ask myself. We disagreed with each other often, but while our opinions were strong, they were weakly held. During the last phase of our research meetings, I would often just sit back and watch them debate each other. That felt nice.

At the same time, I have to praise ACL Rolling Review. Like many of us, we were disappointed with ACL, and perhaps because our dissatisfactions were loud and clear, the review system seems more robust than ever. The reviews I’ve received, and those prepared for others that I see on ARR, tend to be thoughtful and to the point.

Obviously, no system is perfect, and you will still see nonsense here as you would at NeurIPS and places like that, but in my anecdotal experience, I would still very much encourage people to consider submitting there. It feels nice to have your hard work respected, by your peers and ACs alike.

In 2025, I also got to teach my own class at Berkeley!

I’ve long dreamed of teaching my own class. I organized a group study during my first year at Berkeley, which was based on my reading groups at UCL and CMU before that. (fondly remember closing the first session I led at UCL with Ben & Lin-Manuel’s “Found/Tonight”: “tomorrow there will be more of us” -- that was 2018.)

I’ll probably need more space to reflect further on the class, but it turned out to be an applied AI course grounded in literary theory and philosophy. Since many students were taking (or had taken) NLP and/or ML, we covered a dizzying array of methods (see screenshot) and tried to imagine -- while staying critical -- how they might serve as measuring devices to shed light on culture.

Having GSI’d for a few years at this point, I thought I had a pretty well-calibrated expectation of students -- and once again, most of them kept surpassing it! Like what I aspired to in my lectures, in their projects, my students jumped masterfully and convincingly from Walter Benjamin to word2vec, TikTok to CLIP, LLMs to Kant. And one even digitized books themselves to measure memory and violence through KL divergence.

If 2025 was defined by the kind of collaboration that makes you smarter and kinder at once, I can only hope 2026 brings more of the same.