Our paper The Expanded Othello AI Arena: Evaluating Intelligent Systems Through Constrained Adaptation to Unseen Conditions has been accepted to TMLR! Congratulations, Byunghwa Yoo!

This work grew directly out of an assignment from my Artificial General Intelligence course in Spring 2025. Byunghwa did a great job pushing my Othello benchmark with unseen, tweaked rules, and decided to turn it into a full manuscript. (For those who want to try on yourselves: given stages 1–3, can you build an intelligence that solves stages 4–15?)

The paper presents the Arena itself, along with an evolutionary adaptive-minimax algorithm that learns an order of magnitude faster than PPO. The Arena was designed back in April 2025, and while we were preparing the manuscript, the ARC-AGI-3 preview was released concurrently. Since the spirit of our Arena and algorithm turned out to be quite similar to ARC-AGI-3, it was reassuring to see the community moving in the right direction.

I’ll be teaching the Artificial General Intelligence course again this fall, and I’d love to involve more of you. Please reach out if you’d like to (1) join as a TA, (2) help design a new Arena aligned with the future direction of AGI development, or (3) develop your work from last year’s course into a paper. Byunghwa’s project shows that last point is very doable — and I’m glad to support any of these. Let’s build something together!