20k-30k bots on OGS

I began playing go a couple of weeks ago. I am trying to use the five 20k-30k bots to break some bad habits. Della Baduck is a joke - it mechanically fills up one side of the board. Of the other four, Amaranth and Agapanthus are the only ones I can sometimes defeat. The other two seems harder. But the kyu estimates for all of them fluctuate wildly between about 20 and about 30. So what are the “true” kyu strengths of Amaranth and Agapanthus?

1 Like

I’d say that bots at that level have pretty bad habits, so by playing against them you’re likely to pick up more bad habits than when playing against humans of that level.


Weak bots are more likely to teach and reinforce bad habits than break bad habits; I would recommend only playing humans

As to your question, OGS does not allow people to “exploit” the bots’ weaknesses, which is a fuzzy thing (don’t worry about accidentally doing it; you’re fine, I think the rule is more about people trying to rank manipulate), which means that the bots end up inflated relative to their true ranking (imagine if it was against the rules to exploit a weak spot of a human player; really weird quirk of OGS), and also likely explains the fluctuation

1 Like

I am sure you are right! I noticed that the two A’s let me get away with some horribly bad moves without punishing them. But which human would tolerate playing against someone who often even neglects to check for stones in atari, who has a reading depth of 0-1?

1 Like

I suppose other humans of your level make similar mistakes as you do.


We recently got the ranked opponents rank restriction limit lifted, so regardless of your current rank, you can open a challenge for 25k+ to restrict to newer players


That’s good. Thank you for the info! I will try it.


I run amybot-beginner and I use some janky scripts to calibrate it’s strength to hover around 22k.

And wrt to making mistakes, I’m sometimes feeling like I want to say something amybot does must be a bug because it sucks so obviously… But then I see the opponents making similar moves so I end up just confused and frustrated about what is a bug and what is a weakness. And hence the automatic calibration of reading depth (and for amybot-beginner, injection of noise into the evaluation results).


We were all beginners, and more are coming. Relax and enjoy the game.

I too encourage you to try humans they are much more lovely to play.


“Bot exploits” typically involve players improving their rank by exploiting obvious bugs in the bot program. It does not refer to exploitation of the quality of game play in the human sense (by which I mean things like understanding what is big or sente). There are a lot of examples that I am not going to talk about because it would give away secrets, but there is one that I think is safe to mention because it rarely, if ever, happens anymore. Some years ago, many bots, including some strong bots, could not handle ladders. They would play the ladder out to the end and lose, and they would do that in game after game, which some players repeatedly exploited. Human beginners, in contrast, quickly learn not to play out the ladder and not to fall into it in the first place. Of course, a bot with a flaw like that needs to be shut down until the flaw is fixed.


Since I was at 30k a couple of weeks ago, probably a 22k bot is too strong for me right now. Thanks for the info!

1 Like

Thank you for the encouragement!

Thank you for the explanation. I was nervous that the Amaranthus or Agapethus bots would hit me with a ladder, but it never happened.

it may be easier to start vs 25k humans on 9x9 board than vs 20k bots on 19x19.


Humans also would likely not let you play against them 100 times in a row if you keep beating them (or getting beaten!) the same way, but that’s difficult logic to program into a bot.

If bots could easily restrict how often they play against the same opponent, they would be harder to abuse and likely have more stable ranks.


You overestimate 25k humans. Most of them are literally complete beginners or those who don’t understand rules. They just don’t yet able to have ability to be disappointed in moves of their opponent.

1 Like

it may prove challenging to find 25k humans around the clock. humans tend to get stronger or quit, but don’t stay 25k long

1 Like

there are more 25k than 17k:

its possible to accept custom games of needed rank or create your own custom challenge with rank limit or automatch with rank limit. If your account rank is not high, they would not be afraid to play with you.

its possible to have 5k level without any reading. After a lot of practice you can just directly feel which moves are likely correct without imagining any move sequences.

The ladder is the exact sort of exploit I’m talking about. If a player, whether human or bot, cannot read out ladders, they ought to get punished for that and it reflected in their ranking