Tysen Streib

Read some thoughts from me and other experts in board game AI

We wish you a happy Easter weekend and an enjoyable springtime season. We will soon share new WIP Infos. For now join us in our games and make use of a chance to grab them on a discount :) Find out where to buy our games here: digidiced.com/our-portfolio/

@tombos21 BB/100?

Short-stacked cash game strategies are wildly underrated in tough games for three reasons: 1) Shorter stacks mean lower variance. That means fwer less brutal downswings. Most people play worse when after running bad, so this improves your results from a performance standpoint. 2) Better risk-adjusted return improves optionality, and lower variance means you can shot-take higher stakes more aggressively (40bb eff = roughly half the bankroll requirements of 100bb) according to the Kelly criterion. 3) Cash game regs have no idea how to play vs a shortstack. Take them into a deep dark forest where 1+1=3, and the path out is only wide enough for one. 4) If everyone else is deep, over the course of one orbit the lone short-stacked player has a significant theoretical edge. A big stack can be an advantage, but it can also be a liability. Here I measure the GTO winrate of a short-stacked (25bb) player when everyone else is deep (200bb), with 5% rake and a 1bb cap. The shortstack strategy outperforms the full-stacked one by 8.7bb per orbit! Now, practically speaking I think you should play deep in soft games since a lot of your edge comes from fish stacking off too wide. But in tougher lineups there's a real argument to be made for playing short.

There is a Festival on Steam going on that is all about Board Games :) Perfect fit and an additional opportunity to get on "board" ... ^^ store.steampowered.com/search/?develo… #sales #boardgames #indiegame

Is your board game collection at home taking up too much space? Try out those games in digital format, including old classics, new takes on time-honored formats, and everything in between. Steam Board Game Fest is live now, running until February 2! store.steampowered.com/category/board…

With the new update of Caverna we enter the new year! And we are happy to present the winners of the "Weekly Challenge 2025". Well done! The leaderboard is included in the profile and the rewards in form of avatar ornaments are available for the winners. ^^ 🥳👍😃

@tombos21 I’ve been thinking about this recently. I’d love to be able to add a complexity penalty to solutions and then see how things change as you ramp that penalty up. See what edges get shaved off.

I'm convinced that a disproportionate amount of the complexity in GTO solutions are just solvers trying to minimize their blocker weaknesses. GTO strategies are absurdly optimized. Way beyond superhuman limits. I'd guess 95% of the complexity is responsible for reducing the last 1% of exploitability. How much simpler could we make it if we didn't care about this minutiae?

Dark Forest or not O,o... the Winter Sale is going on! Time to grab some games for yourself or as a present for your friends. You can find the store links here: digidiced.com/our-portfolio/ #digidiced #boardgames #WinterSale #Sale #supportindies

@tombos21 I think I just gave 2 players a top 20% hand and checked if the lead player switched. You can do fancier hand weightings but I just wanted to see what came out of this. The mischaracterization of trip boards gave me pause.

@tysenstreib That's an interesting approach. Can you give any specifics as to how you compared similarity of hand orders?

Open poker theory problem: No one has invented canonical flop subsets. We have regular flop subsets, which are a set of "average" flops that are useful for mini aggregate reports. But we don't have canonical subsets that aim to capture strategically distinct textures. For example, if I open the Pio 3-flop subset, you get three unpaired, broadway, FD, flops. Why? Because these are the most common flop types, so they represent a kind of "average flop". They are minimizing the error of the average. What I'm looking for is more like, a representative subset. Something that shows off the different types of textures you'll face. Now obviously I could just vibe my own set of representative flops for each common texture. But I'm wondering if there's a more rigorous way to do it?

@tombos21 Mostly making AI for board game apps.

@tysenstreib What are you doing nowadays Tysen?

A (hypothetical) Bayesian Exploitative Poker Framework: - Priors: Nash equilibrium + pool data - Evidence: Villain info/actions - Posterior: Your best guess of their actual strategy Your goal is to maximize your hand vs their posterior strategy, while hedging towards Nash for uncertainty.

This is easily one of my favorite theoretical 2+2 posts of all time. 14 years ago, the legendary @tysenstreib (co-author of Kill Everyone) dropped an algorithm that solved a problem poker theory had been stuck on for ages: calculating ICM for large tournament fields. Before this, you couldn't compute ICM for more than a few tables. Complexity of the ICM algo scales with the factorial of the number of players O(N!), so it quickly becomes computationally intractable. As a result, large-field MTT strategy was all guesswork. Tysen’s algorithm is simple and elegant: Assign each player a random number, and give them a score = rand^(1/stack). Highest score wins 1st, 2nd highest wins 2nd, etc. Repeat that process many times, and you have your distribution. Remarkably, it converges to ICM precisely in the long run. This humble 2+2 post is the inspiration behind the large-field ICM engines in tools like GTO Wizard and HRC. We use more advanced methods now, but I doubt they'd exist without Tysen's novel insight. forumserver.twoplustwo.com/15/poker-theor…

@JambasketPoker @tombos21 Ideally both. They aren’t incompatible in theory- it’s just that ICM depends on your specific situation, so it’s relevant just for a specific scenario.

@tombos21 @tysenstreib Is it better to play GTO or ICM?

@SavagePoker Allow late starts but remove a full set of blinds every 20 minutes. Easy.

You may not know this but as late as 2003 at the WSOP if you showed up after the tournament started you were not allowed to play.

@castro_rob81631 I’m doing my best. Right now the AI is very good, but not superhuman.

@tysenstreib Amazing! Any plans to make a super-hard AI opponent based on AlphaGo?

Finally the official announcement! So excited for this one!

I’m very excited!

There we go! Pagan: Fate of Roanoke is finally available to play online on Board Game Arena! Will you uncover the Witch hiding among the villagers — or spread your influence before the Hunter finds you? #Pagan #FateOfRoanoke #BoardGameArena #Wyrmgold #Digidiced #BoardGames

Hey everybody, today is the last day of the sale for the Playstore and the App Store! On Steam you still have time until the 10th of July. 🌡️Hope you have a cold glas of water in your reach ;)

It's unbelievable how quickly the year is moving forward. Summer is already here and we're busy, busy... Time for a picnic with great offers!💓🥤🌻🌞 digidiced.com/our-portfolio/ #digidiced #boardgames #SummerSale #sale