If they are truly gto, no strategy you choose will reduce their EV...

http://www.runitonce.com/nlhe/breaking-apart-misconceptions-about-gto/

You cannot beat GTO, otherwise the stategy you've beat wasn't actually a GTO strategy.

If you get a perfect score on a test, then you cannot improve you score.
If you do, then the previous score was not perfect, given the test is taken under the same ruleset.

+1 to comments above, if they are GTO, or close to it, then you won't be able to adjust to beat them.

I once read an article from a guy who said he beat non-human players out of a lot, but they were very much exploitable programmes. He said he invested a lot of time (and money!) finding out what ranges they played in what position, what all their bet sizes meant and what hands they 3 bet with etc. so he could then play profitably against their exploitable play.

I imagine these type of programmes have a random number generator in them for bluffs. So, they'll 3 bet 5% of hands for value, and then 5% of the other hands on a random basis as a bluff. How can you adjust to beat an RNG?

Just to clarify, the program tries to play in a GTO way, but I'm positive it doesn't truly achieve that, or close to it. It has a lot of tendencies that should be massively exploitable such as: It never leads, if it doesn't have the initiative then it won't bet. It opens 92% of hands preflop for a minraise, and always cbets TP+, so it's checking back range is really weak - it's hard to balance correctly when it's range is so wide.

I've been experimenting with a few different betsizes and seeing how it responds, for example when I open raise to 20BB - it's 3bet range is 1010+, A10o+ and it 3bets to 70BB so no fold equity on a 4bet at all - this must be exploitable? it flat calls with 88, 99 KTs/KJs, QJ.

Also, if it 5bets - which it does with TT+ AQo+, then it 5bets Allin, no matter what the effective stacks are..If I MR open, get 3bet to 7BB, I min4bet to 14BB, It'll shove allin 300BB's deep w/ AQo.

I was wondering whether there was any way to tailor my betsize with my hands in order to increase my profit...If I'm playing against thie program and I get AA...Is opening to 2BB..6BB..20BB going to show much of a difference in the profit? I feel this is where I can gain the biggest edge, varying the betsize whereas the program doesnt vary at all...along the same lines, if I have the nuts on teh river, there must be an optimal betsize to bet...a 2BB bet on the river > a 1BB bet on the river, so there must be some sort of way to gain an edge in betsizing, I'm just not sure how best to work it out.

Are there any specific things I could look at postflop in order to determine other spots where it is inbalanced etc?

Thank you for the replies so far!

If he has a completely capped checkback range then you can make arbitrarily large valuebets with any hand that beats his entire range.

For instance you flat AJ preflop and he checks back on KJ83 and the river is a 2, you can now bet 4 times pot for value. The same goes for bluffing in this spot. There might be a betsize in this spot where he just folds everything, for instance if you always bet 12 times the pot in such a spot and you only get called by rivered sets over a very large sample you have a very profitable bluff.

Happy to help someone punish a bot.  Great way to start a thread too.

I'd be interested to see all the postflop stats (Cbet by street, fold to Cbet by street, x/r flop, fold to flop x/r)

Until then, the best way to beat a bot is experimentation with bet sizes and strange lines. Finding something that the programmers neglected to prepare their bot for is definitely the way to have a substantial edge over a reasonably good bot.

No Limit Holdem is a big game, and I would guess today's bots can't handle every single possible line very well. 

May I ask which site has HU bots?

I think the adjustments you have mentioned so far excellent against this kind of bot.  Why not size your open raise pre so as to maximally exploit its defending range?  Or, if not, why not limp a lot pre (or minraise) to take better advantage of these postflop leaks?

no way these bots are GTO,  so exploit away. That's all I have to say.

I think a real GTO bot would also take into account your strategy and calculate most optimal bet sizes. So if you switch your strategy it would switch bet sizes in certain spots. Maybe bet sizes would differ from board to board also. I think because humans don't have time to do it in a game it makes sense to use one-two betsizes for every street (to stay balanced) but a program should be able to calculate the most optimal size. So if he doesn't do it he is likely not a true GTO bot. 

And actually i doubt that any computer can give you a GTO flop strategy because the decision tree is huge.

There have been lots of posts on 2+2 re this...Ben will know far more about these than I do  - from what I can recall the cliff notes are that NLHE is nowhere near solved in a GTO sense; LHE is closer and they are nowhere on PLO...

This is an interesting puzzle, but it doesn't have much to do with GTO. Rather, it's about exploiting a static and probably highly flawed strategy.

"I think a real GTO bot would also take into account your strategy ..."

No, this is contradiction in itself. "Real GTO" doesn´t care about YOUR strategy, it´s fixed, no matter what you do. Please read the link I posted in the entry, tons of discussion about exactly that point. :)

OP said the bot was 'GTO', which he bracketed because it's obviously not GTO.  What he meant is that the bot plays a static strategy which is fairly well balanced and which it expects to win against the player pool with.  The bot might even be using some algorithm to develop its strategy which solves a highly simplified game tree- in this case it's methodology might be GTO, but it's playing a (highly) restricted form of nlhe. 

Here
are some postflop stats:

31k hand sample:

cbet
IP

flop
48%

turn 54%

river 58%

3bet pot cbet OOP

flop
42%

turn 60%

river
79%

Fold to cbet OOP (Just a note, I think that a lot of the bots play is based on Applications of NL Holdem, where it plays in a way in which it calls enough to make your "bluffs" breakeven, so this is highly dependent on the betsizing - fwiw, a little thing I've been doing in an attempt to exploit this is betting big with gutshots etc, as, if called, the bots range is really strong and so there is likely a lot of implied odds if the gutshot hits.)

flop 37%

turn 30%

river
40%

Fold
to cbet IP

flop
31%

turn 24%

river
36%

x/r

flop
10.5%

turn 6%

river
6%

fold
to x/r

flop
42%

turn
23%

river
53%

I
have been testing all sorts of outrageous lines with success for
example:

Calling the turn and mindonking the river with missed draws/hands without showdown value - the bot seems to fold an exploitably large amount in this spot...If my feeling that it is using a "call to make opponents bluffs breakeven model." then I'm not sure how it applies in the sense that, if it's range is really wide, and you minbet, does it call with, say, J high because if it doesn't, that means that you have a profitable bluff? based on this situation, I don't think it does.

Using
minbets to get to showdown for basically whatever price I want as the
bot does not EVER lead without the initiative. A minbet on the flop
and turn is usually sufficient to get to showdown with A/K high or
other weak hands with some showdown value.

Min
3betting its raises as when it calls (with all his value hands
excluding the nuts) it checks the river 100% allowing us to never
face a tough river decision.

Min
cbetting after 3betting on boards 9 high or lower as if it calls its
range is 100% overcards.

Minbet leading after flatting pre
– the bot almost never slowplays (I've seen it do it once or twice,
but for the most part, it should just be a “non 0%”) so when the
bot calls the flop, it's range is essentially extremely capped.

GameTheory:
You bring up a good point with regards to overbetting vs a capped
range. The bot appears to attempt to balance in some spots for
example: It checks after 3betting with air and AA, which of course
means that even though it has attempted to be “balanced”, the
weighting of AA to air is completely skewed towards air in so much
that the balancing doesn't really achieve a lot in that spot. However
there must be spots which he is totally capped (A common spot being
where it will always 4bet JJ+ / Aqs+ so when the 3bet is called, AJ
essentially becomes AK in terms or TP value, and is far easier to
play.) and a range of huge value bets with lots of bluffs prove
fruitfull.

DirtyD:
I am min4betting a polarized range which contains hands that are not
dominated by his 4bet calling range (86o and similar as well as hands
i'm willing to get it in pre with vs his 5bet shoving range) to
capitalize on it 5bet shoving at any stack depth and it is calling
4bets with a super defined range (ATo, ATs, AJo, KQo, KJo). I agree
that this is a fascinating puzzle due to our opponents ranges being
totally static. This is not about "beating GTO" or whatever
but instead beating up on a static strategy.

Ben: I have
constructed my opening size with each part of my range to play well
against its defending range with opening sizes varying from 2 to 20bb
based on my very rough estimations about which calling range is most
profitable to play against. For example: With A9o I open to 5bb; the
largest size where he still calls with all his Axo. I have a
sufficient hand sample (31k) to know its defending ranges vs various
sizes quite well. Limping is one avenue I have yet to explore...The
thing is here, I'm not sure what to take into account, I know the
ranges I'd like to be playing against his defending ranges, and have
developed a strategy accordingly, to make sure that I dominate the
majority of his range in the spots where that is possible...but, I'm
not sure whether that's the main factor to be thinking about in this
situation...In the earlier example, I raise A9o to 5BB, because he
fold A2-A7 when you raise to 6BB...however, as profitable as it's
going to be when he calls and you outkick him and can get three
streets from his aces when the board comes ace high, he is also
folding 58% of the time...and I'm having difficulty working out
whether it's going to be more profitable to win 1BB 58% of the time
and (really complicated formula which works out what your average BB
won is against his whole range) or whether it's more profitable to
use a different sizing.

Again,
thanks for all the responses/discussion. 

Sorry about the formatting, I drafted the post in word, then pasted it into here and the formatting messed up.

Given that its three bet is  fairly constant over all open sizes you should be opening larger with your stronger hands.

Microgaming?

I'd rather not go too much into specifics, as I feel, it helps the programmers. If people play there and are interrested, then I'll PM the names. Possibly notes as well. I've done extreme adjustment vs them without them readjusting at all even though I'm opening up a ton to get exploited. I'm very close to 100% sure that they are bots fwiw. 

I've used a bot when I played Diablo 2 (Video game), about 10 years ago and considering what that did, It's pretty mindboggling to me, some of the mistakes they are making programming wise. Don't want to go into detail here either though, as it will make them harder to detect. 

Good news is that iPoker is taking a stance against the bots it seems. Given their history though, I´d need to see the results before I believe them though.. 

I nearly the exact same stats vs 2x as you do by the way.

The screennames aren't that important, as they switch them..

RE: GTO, bots, is HU solved....for those interested in reading more there is an interesting thread on 2+2, in NVG titled "Heads Up Hold'em Solved?". One of the PhD students from the University of Alberta, "I'm one of the PhD students...that developed Hyperborean and Polaris" chimed in starting at post #43 and made a number of posts after that discussing the current state of the best bots in the world, GTO, etc.  If you are interested in such things, it's a good read.

How is it allowed that these bots play on the same site with humans?