"Though Claudico will run on a CMU computer as it plays the pros, it will use Blacklight during the event to continuously improve its strategy"

took that off the site
https://www.cs.cmu.edu/brains-vs-ai

So looks like the bot is improving as it plays these pros. I thought it was a GTO approximation this bot made using like a neural net like pokersnowie but this bot actually analyzes the rules of the game etc. Maybe not such a great idea for the pros to play it ?

sounds good, but i'd change it from 'massive mistakes' to 'maybe small mistakes'. It should also call down lighter than optimal too if this is the case.

It's just a spot where a huge overbet with the nuts doesn't make sense.

Anyone else's Doug-stream really laggy? At least sound wise? Wasn't like this for me earlier

Yes, and he commented on it earlier so the problem seems to be on his side.

More lines like that! they are mindonking oop almost every opportunity from single raised pots to 4bet pots! I agree it seems to work well. Interestingly it is a strategy I regularly employ in HU SNG vs certain players.

Want to let Claudico know to keep his head up. Challenge going a lot worse than he planned on by this stage but its gonna turn around soon

anyone got an updated lifetime bb/100 for the humans?

Lol, for old-schoolers this.

what do you guys think of the bots play with KK in the pot vs Bjorn. Quite thin

ok i'm sorry but the bot's play with q2 on aa22 shoving the river vs Bjorns minbet. That just has to be bad. OR is it just making this sickest bluff with the quad blocker? can't be right. Bjorn could have AA

edit bjorn had an ace and minbet called AI a huge river overbet

Humans crushing again

Around 1.2k hands into today's grind, scores:
WCG: +10k
Cheet: +26k
Big Dick: -11k
Flushie: +85k

Bjorn just stacked the bot with boat > trips, where Dong didn't get stacked with reverse cards.

dong -6k while bjorn is up 84k is pretty sick. They are the GOAT duplicate poker team lol

yeah i dont know how dong didnt get stacked. If I heard the story he told Doug correctly, he had A4 on 744 vs 77?!

yeah, that's correct.

bots never expect the big dick

Was this bot developed with self play at all? If so how much? How would someone knowledgeable describe how this bot works?

awesome hand

Qd Jh 9s Ts 4c

Bot AQss, WCG AA

Bot 4bets, WCG flats. flop k/k, turn k/b/c, river k/overbetAI (!!)/f

Bot AA, cheet AQs
bot 3bets, cheet calls. hand gets to showdown cheaply

WCG pwned, tho if he 5bets its prob a similar result unless bot derpjams the flop (doubt the bot folds to small 5bet pre)

The best thing about those min leads is that if you play GTO approximation you won't punish them by much you would need to adjust to make those really bad.
I've run a tree similar to that big Ah 4s 3s Qs 8h spot and on the flops evs for lead are like this: http://i.imgur.com/FmKWPXg.png
So the mindonk is never chosen but the difference in ev is often in 0.01bb range (with a spade) and in 0.1 range for hands without a spade. That's very common because it doesn't make sense for GTO strategy to focus on punishing some weird line (it's enough to make it slightly worse). Trying to punish too much would possibly result in some hands being very profitable in that line.
That shows the importance of overall exploitablity. If it's not a good approximation the humans will find a way to get to exploitable branches sooner or later.
Now the question is if it's possible to get approximation good enough in NL or maybe the whole approach won't work. My personal view is that it will work for NLHE but won't for more complicated games.

Almost 14k hands in, Team Human at 10bb/100. Flushie is insulting the bot.

those guys outsides of cheet would win way more than this pre-rake vs most HU bumhunters from 200 to 400nl. Especially 200bb deep.

does jason play online these days? don't know his stars and FT screen names unless it's cheet

I think the right answer here is "maybe" as I'm really not sure and haven't got any samples other than I know that most 6max crushers at midstakes win at 4-6bb/100 and anything over 2bb/100 is considered really good.

AND That still doesn't mean that the bot could beat said bumhunters. Just because "strategy Polk" beats "strategy Bumhunter" for more than "strategy Polk" beats "strategy Bot", doesn't mean that "strategy Bumhunter" would lose to "strategy Bot".

The only thing that is clear here is that the bot is outmatched, which is just a huge failure for Team Bot and really good news for us players.