Question regarding AI reviews for Hane Supporter tier for a beginner

The number of playouts for the lowest Hane Supporter is 1000,
Looking back at the old subscription tiers this is a little better than Dan Supporter which had 800 playouts.
Does this mean that even at the lowest tier in the current pricing scheme OGS provides me with AI reviews that are better than a “Strong professional level player”?

I know that these are the only tiers that are available but as a beginner I can’t help but think this is a bit “excessive”.

Also, from your perspective how good or bad was the 125 playout reviews that I was previously subscribed to?

I think it just depends on how you want to use the AI review.

If you want it to find big points on the board in the opening or middle game, it might do that reasonably well on 1 playout, but naturally better with more.

If you want to analyse some complicated joseki or capturing races or tsumego, it’ll probably help to have a lot more playouts.

If you want to find out why some endgame went wrong or why it was better to ignore/answer an atari it might be fine on low playouts, or it might need a lot if there’s a much bigger concern with a surrounded group, or a related or unrelated life and death problem.

I don’t think I would call a number of playouts excessive or not, from my perspective. More playouts will be better for both (A) deeper analysis by the player and (B) very shallow analysis by the player but trying to be more confident in the score estimation.

I’d guess the 125 playouts one was decent enough. For 9x9 games, they can get surprisingly complicated and it might be the case that the higher playouts for that board size could be more noticeable on average than in say 19x19 games.

1 Like

I believe at the moment there are just two kind of AI:

  • strong AI
  • silly AI

We can’t actually have an AI that plays like a SDK or a low dan or so.
I think that those AI (there’s a bunch also on OGS) are just weak AI that can’t avoid missing opportunities or making mistakes. (I see how this could lead to some discussion about humans making mistakes :slight_smile: )

When you want to analyse a game, you don’t want advice from “noob_bot” (just an example).
I sometimes play against noob_bot but then I review my game with katago.
The more playouts, the better.

The issue with AI is that they don’t explain their moves and their strategies: you just have a score for different moves. Better to have a good estimation than a weak or foolish one.

I think so, but I’m no expert with AI.

2 Likes

I’m not an expert either but from my experience it depends on the game. There are games even among the professionals that a 10 playout AI would win, but there are games where even a 1000 playout AI would lose due to reading mistakes in complicated fights.

I have my katrain configured for fast 100 playout analysis and I still find myself enabling pondering during more complicated positions because it can’t process them correctly.

So yes I would say that the 1000 playout AI is definitely a mid-top pro level in the majority of the games. But it doesn’t mean that 1 playout AI isn’t pro level, it also is, it’d just fall apart during the fights where you need to read deeply (at pro level).

Also why would you think of it as an overkill?

1 Like