Linus vs Nemesis
I have been thinking about human capabitilies vs solver capabilities.
When drilling vs GTOW, even a lowstakes reg will not lose more than like 2-4bb/100 in the GTO trainer or when uploading hands to the analyzer.
I think it's pretty safe to say that Linus would beat a bunch of lowstakes regs for a higher winrate than a GTO Bot would.
But how much would Linus lose vs a max-exploit solver?
Let's say, hypothetically, we connected Linus' brain to a solver and all his best guesses for GTO frequencies and sizings
for all combos on all boards and runouts where uploaded to the solver. The solver than max-exploited Linus' entire strategy.
How many bb/100 would Linus lose to the Nemesis Solver?
7 Replies
If I'm not mistaken, Slumbot lost to RuseAI (gtow) by 12 bb/100, and Slumbot is a bot, so more than that I would guess, way more.
Although the bots were playing deep. Still think it wouldn't be less than that.
I have been thinking about human capabitilies vs solver capabilities.
When drilling vs GTOW, even a lowstakes reg will not lose more than like 2-4bb/100 in the GTO trainer or when uploading hands to the analyzer.
I think it's pretty safe to say that Linus would beat a bunch of lowstakes regs for a higher winrate than a GTO Bot would.
But how much would Linus lose vs a max-exploit solver?
Let's say, hypothetically, we connected Linus' brain to a solver and all his best guesses for GTO frequencies and
There are a couple of problems with the question
The nemesis can't exist since linus' strategy would not stay fixed, the computer would have to somehow gather information from linus' strategy the regular way and then build a counter strategy. Winrate will depend on how well it does all this which is not a theoretical problem as much as a practical and logistical one.
If the "machine" actually magically predicted how linus will play each hand in range on every individual spot then winrate would be absurdly high, something we've never seen. It would probably matter little whethere its linus or some other player
There are a couple of problems with the question
The nemesis can't exist since linus' strategy would not stay fixed.
For the sake of discussion, let's assume static strategies, since dynamic strategies makes things infinitely more complicated.
Linus' best guess for what GTO looks for every combo on every board and every runout is magically input into the solver once. What is the exploitability of this static strategy?
(Let's say Linus' accuracy is as if he had 20 seconds to think about each combo, using only his mind)
I guess it also matters a lot how many bet-sizes are allowed.
Linus would presumable only want to use one size per node no matter how many sizes are allowed.
Nemesis would want to use the maximum and would probably also want to use some obscure sizings that linus has never studied.
But at first run, say Nemesis is only allowed one normal/GTO size per node.
Even assuming static strategies is a misnomer. Linus is, as far as I'm aware, still an actual human being, so whether he bets 33% frequency or 66% in a given spot is going to fluctuate/depend on whether he had eggs for breakfast or not.
Eggs -> 66%
No eggs -> 33%
But if the solver does know with 100% accuracy what Linus had for breakfast, then yeah the guy is toast.
For the sake of discussion, let's assume static strategies, since dynamic strategies makes things infinitely more complicated.
Linus' best guess for what GTO looks for every combo on every board and every runout is magically input into the solver once. What is the exploitability of this static strategy?
(Let's say Linus' accuracy is as if he had 20 seconds to think about each combo, using only his mind)
Extreme loss rate. Maybe >30bb 100. The main source of winrate would be how every slight imbalance on river would be met with an extreme and perfect deviation.
But again its a purely hypothetical scenario since you can't physically know how your opponent plays every hand in range
