For the Fall of 2025 - I took Knowledge-Based AI (KBAI)

I wanted to take this class since it is taught by Dr David Joyner. He was one of the earliest people involved with starting the OMSCS program, and one of the few professors still involved since the beginning over a decade ago.

The course focused (fairly abstractly) on the architecture of AI agents. The lectures discussed semantic networks, planning, the SOAR Architecture, and other topics.

There were alot of assignments, projects, and reports due throughout the class. The most interesting was the semester project. We worked on a program which can solve a subset of problems in the ARC AGI Challenge. This is a series of visual puzzles which humans do fairly well at (a human panel gets >98%) and current LLMs and statistically based Machine Learning methods don’t (for instance Deepseek R1 gets around 15%). Here is the leaderboard. While my end result was not groundbreaking, it was very interesting following along and reading about new state of the art models as they came out.

Ultimately - and ironically - those statistical models performed better and better as the semester went on. Gemini 3 Deep Think got 87.5% right on ARC-AGI-1 (which we were focusing on during the class).

This class has made me more excited to move forwards with the statistical machine learning classes (e.g. Machine Learning, Reinforcement Learning, and Deep Learning) since that really seems to be where meaningful research progress will continue to be made.