^ that one was sick awesome

Major respect for cheet, bjorn didn't really want him to call either but he stuck to his guns.

doug's video is down?

Yeah I'm only seeing cheets stream as working right now

Sick it did that bluff w/o any spades. Would have thought blockers would come in play more but cheet said it just airballs at times. It did have T5.

What was the whole hand?

Ah4s3s Qs 8h

Cheet As2h
computer Tc5d

comp raise, cheet call,
cheet minleads flop, comp calls,
cheet bets turn ~500, comp calls,
cheet checks, comp jam 19k into 1700ish, cheet tanks forever and calls

Damn...

I can sorta understand bot's logic tho. If we are shoving the nut flush otr as well as shoving other bluffs there is no point shoving with a lone flush blocker unless it's the As. But with the As there is also no need to shove since either nai bet or check is better.

With s blocker it makes sense to make nai river bet as a bluff instead of ai. The only problem is the river bluff could just be -ev in a vacuum with all hands since bb's range could be uncapped.

Actually it should be just about OK. BB doesn't have enough nut flushes in his range anyway, and not that many lone As that's not 2 pair on the flop.

And risk reward is like 19125/1750 so I can totally see BB folding over around 90% of the time without As.

turn action wrong. He checked Claud bet 525.

^oh yeah, thats right

The bot really puts the NL in NLHE.

Rough session for big dick. Shows how swingy it can be playing vs a bot who shoves 200bb all the time lol

Of course it is. Any spade removes a combination of AsXs allowing it shove more hands given it polarizes between air and nutflush.

Technical difficulties delaying our restart

sry, who is cheet?

lol, hi cheet^, who are you?

Jason

Yeah but you can argue that single spade blockers can have higher EV bluffing with a smaller sizing- or at least indifferent to small size or large size. Also a lot of suited Ax are 3bets preflop as well.

But with a pure airball like T5o the only size it can be bluffed at GTO is all in. So this play works as long as there is a large enough range disparity.

Do you think a bluff here in position should have zero EV? If T5o is a GTO bluff here it must be the indifference hand imo.

claudico confirmed mahatma

i think claudico's aggressive river betsizings are awesome but its river showdowns via bluffcatching feel weird

it's cool how the bot independently concluded that some of these river overbet spots are closest to GTO

It's quite obvious I think that T5o without a spade isn't the best combo to bluff all-in there.
Now, the way those selfplaying algorithms work is that they don't have any "poker logic", they understand the rules and just adjust to what is currently more profitable. It in turn means they probably don't have very good accuracy (not surprsing as NL trees with many bet sizes are humongous). For the river solution though though they have a lot of seconds to recalc it on 64 cores which should be enough to get to almost perfect equilibrium even with like 10 bet sizes. I have trouble imagining how T5o no spade could end up being an all-in bluff there.
Again, it's very strange from my perspective, seems like something is wrong in the design.
Maybe that minber OOP really works wonders there. It's probably a very rare action in GTO (if ever taken) and the bot may have problems estimating opponents range properly. If it thinks that range is very close to 0 then some things connected to computer number representation might come into play and if it uses sampling strategy which depends on how often stuff happen then maybe those branches are not very well calculated.
Anywya, it seems donking OOP is a good idea, it can't possibly cost much but there are many ways things could go wrong for the bot.
Maybe the humans should try more lines like that.

Without knowing the computer's exact range in this spot, can we really say that there are better bluffing hands? In other words, he might not have the hands you are thinking of in his range at this point. Or there might be some better bluff hands (ones with spade) that he bluffs 100% , but he still might need to add a few more worse ones to get to the right value/bluff ratio for that particular bet size

i dont think we need to know it's exact range to make the assumption that it has some trash/spades on teh river if it gets there with T5o

I do think its the min flop donk that is ****ing it.


It's hard to think of some exploitable consequences of this mistake. I guess it means that not accurately estimating a villains value range on the river, so it won't be taking blockers into consideration as it should be. Which means the guys should have some slight hope with river decisions and blockers.


Or maybe it just has some balanced spaz built in to throw people off...

We can't be sure because nobody approximated the whole NL yet. We do have quite decent approximations though for games with less than all bet sizes (say 2). In my experience at least the river ranges in that line should always contain enough combos with spades that you can't afford to bluff all of them, let alone add stuff without spades.
Claudico team for sure has answers, it would be nice to get some insight. I see two possibilities:

a)the approximated river range was such that T5 no spade turned out to be a bluff (as there wasn't enough better hands to bluff in there)
b)the river wasn't approximated well enough and the all-in frequency for T5off didn't converge close enough to 0 yet

I have problems believing a) is possible. On the other hand b) seems even less likely if they recalc it on 64 cores every time (because rivers games are small and that's a lot of cores run for a lot of seconds).
Maybe we could get river ranges from Claudico for that spot and see which one is true.

If hypothesis about rivers not being solved well enough is true then probably value shoving a lot should bring nice profit as there will still be a lot of random calls which didn't have enough time to converge to 0. WCG was already paid off a few times like that.

So the way to exploit it- keep pots small until river with mindonks/bets, so that the bot assigns weird ranges, and makes massive mistakes with overbets?