A Machine Learning Researcher's Notes from 5k Hours of Tekken
2•taha_moji•1h ago
Author here. MS in CS from Georgia Tech. Published at NeurIPS and NAACL. Spent ~5,000 hours on Tekken and reached top 0.5% globally.
This is what I found.
At 60fps with 50ms decision windows, you can't react. You have to predict. Your brain compresses 900+ possible moves into threat categories, reads patterns from partial information, and updates when you get punched in the face. Repeatedly.
This hit #1 on r/ArtificialIntelligence today. 130K views. 93% upvote rate. Apparently people wanted to read about fighting games and cognition. Who knew.
Happy to discuss the methodology or why CPU opponents that read your inputs aren't actually "solving" the game.
https://www.reddit.com/r/ArtificialInteligence/comments/1pq7cnw/what_5000_hours_of_mastering_tekken_taught_me/