Will take humans to win the challenge (to win = being up money). My 3k to your 4k. Pm me if interested. Will find reputable escrow.

+1

Day 6 results:



1.81bb/100, humans up ~110k for the day, Dong leading the comback with +80k let's gooo

There's probably some merit to trying to figure out a threshold where you can capitalize entire pot preflop with a worst card (offsuit 2) in your hand by having an ace blocker. Just my guess though.


This hand is sooo good example why NL>PLO, when QQ works as such a dynamic bluffcatcher. It's like QQ is x/c (assuming no x/r is allowed) versus some very small x% PSB. Then fold vs all sizes between x% and y%, and then again x/c versus sizings bigger than y%. (And then probably fold again after some insanely high all-in sizing where villain only valuebets KQ, not even Q9, and bluffs QQ(+KK), and KK becomes a better bluffcatcher.)

If we take into account the 58k won by brains that didn't count, brains are already leading now.

Laudico is a ****ing animal

But this logic is flawed. If humans out of the gate are playing closer to optimal then the bot stats don't matter.

I'm not saying the humans aren't good though.

It seems more complete to say the player(s) who are not as good should perform less bad as time goes on as they learn from the better player(s). Or, one side was better all along. Either way, the logic behind total carnage doesn't make sense given this. Would like to hear more of your thoughts though.

Lol, bot got wrecked today.

This is weird, how is 5b/c better than just calling the 4b? Seems a little loose 200 deep

Is it just because they think the bot's technical game in 4b pots is better this deep and therefore would rather 5b/c to take the equity chop rather than play it out?

Ot they have enough sample to think the bot 6b bluffs too much? Dunno if even 20k hands is big enough sample for that, this spot is so rare

bot is an animal

I think I actually misworded my argument slightly when I said disadvantage. I do not think the humans have a disadvantage here in this match. I meant that they have a severe disadvantage on their day 1 winrate compared to even days 2 or 3, because playing blind against an opponent is far less profitable.

I'm not sure how familiar you guys are with HUNL, but at a top level all matches are online and with HUD's and stats. This is critical to the adjustment/counter adjustment game, as it offers insight in the array of ways that your opponent thinks about the game. Their stats influence open, 3b, 4b, and a variety of postflop decisions and strategy. I have played almost none of the hands in my career without this information.

I think the fact the humans do not have a HUD greatly hurts them initially, because they cannot make simple adjustments out of the gate such as preflop frequencies and comparing the value on non equity bluff lines.

If we look at the human results by day it's something like

-75k 3k hands day 1
-75k 6k hands day 2
-40k 6k hands day 3
+140k 12k hands days 4-6

I think this just highlights my point as they can prepare more specifically for their opponents their win rate should increase. The humans advantage is being able to pinpoint weaknesses and to exploit them to create edge, without stats that would be very hard to do.

Humans are notoriously bad at estimating percentages. Try to play 1.5k hands in a day and then list opponent x/r river frequency. It could be 9% or 14% and feel the same, but those are massive differences in the way you should combat it.

I have a bit of an opposite argument on stats, I think they matter almost immediately. If someone plays 10 SB's and opens 8 of them, while their true open is almost never 80%, it is in fact your best estimate. So I would use stats almost immediately out of the gate.

Feel doesn't "kick-in" in the way you describe. HUNL is about precision. I am a scientist not an artist.

Within the constraints of the bots bet sizes, the humans are almost certainly worse. If they chose to use only sizes the bot understood completely, they would most likely get crushed.

Day 1 you don't know the full array of sizes the bot will use. This means you do not yet know what sizes will be optimal against the bot.

I imagine that humans play in general much further away from optimal than the computer opponent. However, they are much better at adapting and creating counter strategies. If the humans just played every hand to their best ability as to how they would normally play it, it is unlikely they would stand a chance.

The problem for the computer is that because it doesn't play GTO, it has weaknesses. And that is where humans can shine, figure those weaknesses out and create winrate. As the bot gets better, those leaks are harder to find. So as the bot gets stronger, stats become more important than ever before.

Was a hud used in 2015?

Dont know if it's been posted before but Alex Sutherland wrote a nice article about the bot's exploitability.

go on humans


yall are animals

---

edit also the article posted above is interesting

BDDK,please tame this ****ing beast, us humanoids depend on you. Grab it right by the pussy!

I think there were 2 spots where u pussied out otr (T4 on J325K, K3 on JTx3A) and raised very small to have to bot snap call. In the same spots the bot went all in with T4 and checkraised massive with K3....

That's more likely than it being an info thing probably, but you'd need to know the whole range/frequency to do the math. I could have a 3bet/fold range and a 3bet/call range, but you wouldn't konw what those ranges are if you only see 1 hand.

If I 3bet/fold J2s, you could use math to show that's -ev because not enough fold eq, but if you look at how it fits into the whole range including the 3bet/calls then it would be better than not having it even though its -ev on its own.

He may be 34x/fold with some hands and 34x/call with others, you can't say its bad if you only see 1 hand. I would give him benefit of doubt that it was well thought out, esp since it has an ace blocker with weak kicker which seems like it could fit into that range well


I was a bit of a douche too

This entire thing is so bad-ass. Oh humanity. Good job AI team, good job casino, good job sponsors, and good job team brains.

After railing this for way more hours than any sane person would do (time flies, and Jason Les is entertaining) I can only echo what Dong wrote: "bot is an animal"

Seriously. The bot is ****ing bad-ass (as bad-ass as a non-feeling entity can be). I mean it 5x pot cb bluffs wet boards. When did you last time see someone 6x pot the turn against world elite players and not get crushed? Snaps off river jams with tpgk on wet boards, etc etc. Mad.

Made a tiny mistake on the first graph, forgot to divide by 4 the number of hands for each player.



Nice come back from Dong ! Seems like there is hope for humanity

But isn't that what the machines would want us to think?
This has a Turkish-coup feel to it.

Even though the interface is beyond terrible and hurts my eyes, this has been fun to sweat. Cheet is doing a great job interacting with the chat

That guy Jason Les was really good in the booth during the wsop too.

Just an all round likable guy.

Big thanks to everyone posting hands esp with the mirrors. This is fascinating


Sent from my iPhone using Tapatalk