KataGo
2 stories filed under KataGo on Beyond Market Intelligence. The newest of them: “LLMs prove strategic reasoning on Go, with room to grow.” and “Exploring How Neural Nets Learn Go's Hidden Symmetries”. GoBench puts large language models to a properly hard test: 9x9 Go against a ladder of KataGo opponents, from random to superhuman. KataGo's models aren't told to respect Go's rotational symmetry, yet they might learn to anyway. Acme AI is the next-generation, AI-powered spreadsheet platform built to replace Excel and redefine how analysts, data scientists, and enterprise teams work… The list below is every KataGo story on Beyond Market Intelligence, newest first.
LLMs prove strategic reasoning on Go, with room to grow.
GoBench puts large language models to a properly hard test: 9x9 Go against a ladder of KataGo opponents, from random to superhuman. The results are telling. GPT-6 Astra max lands at 2500 Elo, while the best KataGo sits at 4400, and even with coding tools and two hours of prep, Codex with Astra reaches only 3560. That gap matters. The benchmark's strong correlation with ARC-AGI 2 (r=0.83) suggests it is measuring genuine reasoning, not memorization. It is a refreshingly unsaturated evaluation.
Exploring How Neural Nets Learn Go's Hidden Symmetries
KataGo's models aren't told to respect Go's rotational symmetry, yet they might learn to anyway. This study asks a sharp question: do superhuman networks internalize orientation-independent concepts, or do they quietly memorize each rotated view separately? The answer, it turns out, surprised the author. For anyone curious about what neural nets actually do under the hood, this is a gentle, honest look. It's also worth noting how much AI assisted the writeup. If you enjoy this, "Unlock LLM Training" offers a related practical angle.