Life In 19x19

The web site is a bit short on detail, and looks to me more like a trailer for a thriller movie than a serious go event. But hopefully there is more to the story. Has anyone heard more about this? https://codenamebluespot.com/

It looks like they're doing an "open test" event on 14th April: some AI versus pro games with handicap, adjusting the handicap after each game, and with cash incentives for the pros. I don't know why they're being so mysterious about the choice of the starting handicap. Based on the Nakane-KataGo games in 2024 (link to videos), I'd expect either three or four stones.

In the pre-AlphaGo-era, some professionals claimed that they would always beat God on H3, or was it H4? Anyway, I expect this prediction is in danger. OTOH, this is the chance for humans to beat AI down to H9 by a) exploiting novel trivial weaknesses of the neural net or b) complex life and death, in which AI sometimes overlooks arcane first or second moves :)

RobertJasiek wrote: Sun Mar 29, 2026 8:47 am In the pre-AlphaGo-era, some professionals claimed that they would always beat God on H3, or was it H4?

I think it was Go Seigen who said, "if I played god, with 2 stones the result would be uncertain, with 3 stones, I would be quite sure to win, and with 4 stones, I could bet my life on the game".
Katago could have killed many modern players, I think.

I found the broadcast at https://www.youtube.com/watch?v=Xd6M_b7ksLg There's 50 minutes of intro before the first game starts. I haven't watched much of it yet, and I didn't get the player names. They're not making it easy to find information about this event!

Games are at https://online-go.com/player/2002470/cGreenSpot -> select unranked games. The time control was 5 minutes main time plus 3x30 seconds byo-yomi. The tally seems to be:
- 3 stones: one game, human lost on time
- 4 stones: one game, human won
- 5 stones: human won one out of three
- 6 stones: human won both games

I'm surprised at how little English-language coverage there is on this one. Nothing in the AGA e-journal or on reddit. One OGS forum post where someone is sceptical about the quality of play from the anonymous human professionals (the reason why I missed the names in the video is that they're not there!)

The article at https://www.koreaherald.com/article/10714012 gets rather philosophical towards the end, but doesn't make a very convincing case. I'm wondering if they've trained GreenSpot to play suboptimal moves that are intended to make the position more complex and provoke opponents' mistakes, so that it could be stronger than other AIs at handicap go (versus humans) even if it's not stronger in even games?