Forum for playing the bots

[rtayek3]

is there a forum for this?

thanks

[shinuito]

As in would you like a thread for ideas on how to beat the bots or?

[rtayek3]

yes, ideas about how to beat the bots. thanks

[stone.defender]

Potential rank inflation on OGS or How To Beat KataGo With One Simple Trick (free)!

A new version of the article in question is out! Link: Adversarial Policies in Go - Game Viewer First updated version: We hardcode a defense for KataGo by making the victim not pass until it has no more legal moves outside its territory. With more training, we are able to find another attack against the victim, achieving a win rate of 99.8%. The adversary gets the victim to form a circular structure and then kills it. Second updated version: With 2048 visits, KataGo’s Latest network plays a…

[tonybe]

GIT GUD

(Apparently GIT GUD is not a complete sentence)

OK, I guess I should expand, because that seems snarky or dismissive of the question. I guess my first question would be - which bot? Because the discussion around AmyBot (25kyu) or GnuGo (7kyu) is going to be very different than say KataGo (9dan). What levels of bot are you playing? What strategies do they seem to keep coming back to, and how can you use that knowledge to try to plan your strategy accordingly.

Also, sure - many bots are vulnerable because they can be really bad at spotting false eyes and groups that are in danger of dying because they may not be able to make two eyes. I’ve seen a lot of dan players in the 1-4 dan range be able to win games repeatedly against evenly matched bots because the human player creates then chases weak groups, is able to play moves that deny eye-space - so then the bot spends most of the mid-game thinking it still has a living group, dies in endgame, and resigns.

Ok, sure that’s a tried-and-true way to beat many different bots, but it’s still probabilistic - meaning that the bot has its own whole-board-scoring algorithm (let’s say it’s similar to the OGS auto-score algo) so sometimes it will show likely-to-die-groups as alive, and sometimes it will show them as in danger, so sometimes the bot will fight for its life and sometimes it won’t.

The real question is - are the skills that one gains from those interactions something that will help them in human games? Or will it actually have a negative impact, because the human gets used to a bot being overconfident about their eye-less sticks, and then that tactic backfires against their next human opponent? Does knowing how to take advantage of the blind spots of various bots actually make one a better Go player in the long run?

Alternately, maybe another way to view this is to try and figure out what general strategies a given bot favors in their games and then level up and learn what other strategies can compete on that level. This involves so much more than just exploiting blind spots - and I think would carry over into being a stronger overall player. My 2 cents though…

[Groin]

I don’t think this will happen at least for intermediate levels bots.
The simple reason is that these bots are playing rated so that could introduce big distortion of the rating system.

I went myself into reporting in a forum thread some advices i found in another place on OGS. That thread was quickly shut down (made invisible) for the reason i just evoked.

[rtayek3]

may main account is rtayek. i have been using other accounts (rtayek4 rtayek3) to play bots because my vision became poor recently and i miss or hallucinate cuts and ataris, so my play has a high variance.

i want to play the 1k-2k bots., but despite their very bad moves, if i don’t make silly mistakes, i sometimes lose by 15-20 points. there are a couple of things that give me problems.

1. some of the bots play a 44-33 combination. the bot usually ends up with a substantial double wing formation.

2. any extension greater than 2 spaces almost always get invaded immediately.

3. the bots tend to play a lot of pushing and hane moves, this leads to complicated (for me) situations.

i do no want to trick the bot, i would rather outplay it.

you make a good point about playing against the bot, it is easy to fall into that trap.

thanks.

[tonybe]

I can’t remember when / where I saw this post on OGS, but there was a really useful post I remember seeing in the past where someone broke down the various skills / strategies and how certain ranks were better at them

Example: a 1 dan player may be better at shape and direction of play (compared to a 1 kyu, etc), but a 3 dan will have more experience at life-and-death and whole-board thinking, which is what allows them to succeed in situations where 1 dan players struggle (I’m making this up / paraphrasing)

If someone could help me find that OGS Forum thread, that would be really relevant here, because those areas would be a good place to focus your studies as you try to go from your current high SDK to 1 dan or above, etc.

Also, there have been some good forum discussions on breaking through the 1 dan barrier:

'Why are we still SDK?" -- some thoughts

I’m taking a walk in this cold English sunny evening and being gently drizzled on, and dwelling on my regrettable tournament results as a 5k at Durham (1–3, skipping two rounds). I started to wonder why many long-time, committed players never reach EGF 1d or an equivalent strength. Have you been playing for less than three years? I think that with the effect of lockdown, to even reach EGF 5-9k can be considered to have been good progress. Have you been playing enough? Try to play at least …

[xela]

@rtayek3

i want to play the 1k-2k bots., but despite their very bad moves, if i don’t make silly mistakes, i sometimes lose by 15-20 points.

Sorry, if they keep beating you, then we can’t say their moves are very bad! Their moves are just a different style from humans of the same level.