|
silvergoose posted:I'm mostly poking fun at the fans who say things like "who cares about balance except tournament style players who can go find a different game" who are also saying this is a great thing. My favorite are the people who admitted they don’t play competently and loudly talk about how won’t need the balance changes. Now I haven’t played it but an average score differential of +80 points for futurists vs the lowest means you’ve gotta be playing pretty suboptimally.
|
# ? Nov 22, 2019 19:07 |
|
|
# ? May 11, 2024 15:51 |
|
i'm the epic game reviewer who plays a game one time and has a very good handle on the game dynamics based on a single play with equally clueless players yes, i've never found a problem with the halifax hammer
|
# ? Nov 22, 2019 19:21 |
|
Megasabin posted:I have no interest in Stonemaier games after Scythe, but this seems like a good thing? He's taken a step back and actually admitted there's balance issues and is attempting to address them? Seems fruitless to call him out on hypocrisy when he is making a positive change. People should be encouraged when they pivot in the correct direction. They openly admit they don’t really know how good the balance changes are. Multiple changes are lazy victory point adjustments that don’t address the playability of the civ, and it’s pretty crazy that one civ has to start with up to a whopping 45 free VP just to be “balanced.” It all comes off as rushed and amateurish.
|
# ? Nov 22, 2019 22:12 |
|
Yeah giving someone a quarter of a decent end game score is a hilarious bit of balancing. Also it's good to make fun of him so that he actually spends time developing his next game.
|
# ? Nov 22, 2019 22:28 |
|
Just have Tau-Ceti Deichmann create an algorithm to test your game a thousand times for balance. Couldn't be simpler.
pospysyl fucked around with this message at 22:47 on Nov 22, 2019 |
# ? Nov 22, 2019 22:44 |
|
pospysyl posted:Just have Tau-Ceti Deichmann create an algorithm to test your game a thousand times for balance. Couldn't be simpler. Developing an AI for a board game is a fair bit of work right now, and often the resulting AIs will still miss broad swaths of strategy, or be vulnerable to simple exploits. But I expect that to change a lot over the next 5 or 10 years. I think we'll see AI frameworks that can go from base rules to strong agents quickly and without much interaction. To be clear, a game balanced for AIs might not always be balanced for humans - but it'll give designers some powerful tools for exploring possibilities.
|
# ? Nov 22, 2019 23:17 |
|
I've been thinking it would be very funny to train one of those Go or Chess AIs on the dumb bullshit we play instead.
|
# ? Nov 23, 2019 00:16 |
|
Mr. Squishy posted:I've been thinking it would be very funny to train one of those Go or Chess AIs on the dumb bullshit we play instead. Someone did once: http://keldon.net/rftg/ Race for the Galaxy is pretty much custom designed to train neural nets for and have them wind up much, much better than human players, and... well, someone did. There are a LOT of games this is a lot less true for, especially games where only a subset of components are used in each game, since you can't then just chuck the AI the whole component and ruleset and let it play itself forever anywhere near as easily. But even then, I'd be pretty surprised if you couldn't make AI players better than human players in most games. But conversely, it's aso difficult to make them abel to track the board state etc in a hardcopy game, and making hardocpy AI good is a whoooole nother level of difficult.
|
# ? Nov 23, 2019 00:21 |
|
Holy poo poo, someone made essentially a roll and write for Magic Realm called Magic Realm Lite 30 that I only just discovered. There's even an expansion: https://boardgamegeek.com/boardgameexpansion/194882/magic-realm-light-30-asterisk-expansion The idea is to condense the different systems in MR into sort of mini games which can serve to introduce players into the ideas and systems in the main game. Neat.
|
# ? Nov 23, 2019 01:14 |
|
Mr. Squishy posted:I've been thinking it would be very funny to train one of those Go or Chess AIs on the dumb bullshit we play instead. The Terra Mystica digital developer tried this approach to disappointing results. (If he just had AlphaZero's equipment, it probably would have been better but still these things usually scale such that throwing power at it doesn't get you where you want). https://digidiced.com/2018/designer-diary-search-alphamystica/ No details yet, but he said he's had better success with a learning AI for Viticulture.
|
# ? Nov 23, 2019 05:07 |
|
Mr. Squishy posted:I've been thinking it would be very funny to train one of those Go or Chess AIs on the dumb bullshit we play instead. Isn't this what the Race for the Galaxy app is? That app's AI is brutal, and I thought I read once that it found crazy strategies that no one had found before.
|
# ? Nov 23, 2019 11:19 |
|
rchandra posted:The Terra Mystica digital developer tried this approach to disappointing results. (If he just had AlphaZero's equipment, it probably would have been better but still these things usually scale such that throwing power at it doesn't get you where you want). Yeah. It's still easy to train an AI in one specific task, but as that task becomes more generalized it becomes exponentially more expensive (in terms of time and processing). LIke that Starcraft AI that was built: * Could only play protoss. * Only works AGAINST protoss. * Was built to defeat a specific pro player. * Cheated on how it interacted with the game (It could monitor and issue commands faster than it physically possible in the UI)
|
# ? Nov 23, 2019 16:29 |
|
They eventually did make a Starcraft AI that could play Starcraft II on the public ladder at grandmaster level. But that took them like two more years of Google-tier resources after the limited Starcraft AI. Even then it still has issues with inhuman micro. The new version has a limiter that keeps it at pro human levels of effective actions per minute. But the AI never misclicks and it knows it cannot misclick.
|
# ? Nov 23, 2019 16:56 |
|
homullus posted:I think Kemet is my favorite dunes-on-a-map game. Dune is my favourite dunes on a map game
|
# ? Nov 24, 2019 07:28 |
|
Just played Trismegistus at BGG Con, and wow, it does not seem like a game that can be really figured out after one or two playthroughs - the rules are complex, but the difficulty for me was more around it being difficult to discern what future series of actions was ideal and then backtracking to the current turn.
|
# ? Nov 24, 2019 08:04 |
|
golden bubble posted:They eventually did make a Starcraft AI that could play Starcraft II on the public ladder at grandmaster level. But that took them like two more years of Google-tier resources after the limited Starcraft AI. Even then it still has issues with inhuman micro. The new version has a limiter that keeps it at pro human levels of effective actions per minute. But the AI never misclicks and it knows it cannot misclick. The most recent version recently beat the best SC2 player 4 out of 5, with 2 out of the 3 races. It also has the big bonus of being able to effectively micro in multiple locations at once, with perfect knowledge (within its visible area). So it can always be split harassing, while a human player will never be able to get it off guard with harassment.
|
# ? Nov 24, 2019 14:01 |
Wafflecopper posted:Dune is my favourite dunes on a map game Thirsty Twerps Deluxe has to be mine, I think.
|
|
# ? Nov 24, 2019 14:22 |
|
silvergoose posted:Thirsty Twerps Deluxe has to be mine, I think. second favorite is Targi
|
# ? Nov 24, 2019 14:55 |
|
Wafflecopper posted:Dune is my favourite dunes on a map game Gotta be Camel Up.
|
# ? Nov 24, 2019 15:06 |
Chill la Chill posted:second favorite is Targi Targi is still on my want to play list
|
|
# ? Nov 24, 2019 15:32 |
|
silvergoose posted:Targi is still on my want to play list You should play it, it's very good. Probably me and my wife's two player only game.
|
# ? Nov 24, 2019 15:44 |
|
I played a game at lunch the other day and I'm trying to find a copy of it to bring to Thanksgiving. Maybe it will sound familiar to someone here: It is a card game where the goal is to score the least points in one or several rounds. Everyone gets a hand of ~8 cards (not sure the exact numbers) The cards are numbers from 1-100+ (not sure how high it went because we didn't use all the cards. Each card has some symbols on it to represent how many points it is worth. Four cards are dealt face up and each is the start of a row. Every player will play a card from their hand face down in front of them. When everyone has played a card, you flip them over. Cards are placed in the rows in ascending order. Cards must be played in the lowest possible row as long as they are higher than the cards already in the row. The player that adds the fifth card to a row must take the row and get all the associated points. Does this sound familiar to anyone?
|
# ? Nov 24, 2019 19:43 |
|
Can't be anything but 6 Nimmt! https://www.amazon.com/AMIGO-4910-6-Nimmt/dp/B00006YYXG/ref=sr_1_1?keywords=Nimmt&qid=1574621131&sr=8-1
|
# ? Nov 24, 2019 19:47 |
|
That is indeed it. Thanks.
|
# ? Nov 24, 2019 19:56 |
|
The biggest takeaway from StarCraft machine player development isn't that they can react faster and more accurately than any human, it's that rule-based AIs (meaning a state machine, a fancy flowchart) can always be exploited in one way or another. But a machine that plays dynamically and heuristically, allowing for emergent behaviors, is a different opponent entirely.
|
# ? Nov 24, 2019 20:23 |
|
The only winning move is not to play.
|
# ? Nov 24, 2019 20:28 |
|
S.J. posted:The only winning move is not to play. Also to beat the opponent, that's a pretty good winning move.
|
# ? Nov 24, 2019 21:44 |
|
Ai’s are wimps just unplug their cable GG \/ they can't nuke us if they're not connected to anything 🤖 Bottom Liner fucked around with this message at 23:09 on Nov 24, 2019 |
# ? Nov 24, 2019 22:47 |
|
Bottom Liner posted:Ai’s are wimps just unplug their cable GG You want Judgement Day? Because that's how you get Judgement Day
|
# ? Nov 24, 2019 22:53 |
|
If anyone sees Food Chain Magnate Ketchup in stock online (CSI/MM/Gamenerdz) please post here and give me a heads up.
|
# ? Nov 24, 2019 23:14 |
|
My friend ordered it from an Indonesian board game shop, which is awesome for many reasons, and it arrived in less than a week. He didn't make it sound like cost + shipping was really expensive so I'll ask him what store it was.
|
# ? Nov 25, 2019 00:54 |
|
I forgot about Targi for my dunes-on-a-map joke. Targi is definitely my favorite--that is a clever game!
|
# ? Nov 25, 2019 01:15 |
|
The Eyes Have It posted:The biggest takeaway from StarCraft machine player development isn't that they can react faster and more accurately than any human, it's that rule-based AIs (meaning a state machine, a fancy flowchart) can always be exploited in one way or another. But a machine that plays dynamically and heuristically, allowing for emergent behaviors, is a different opponent entirely. Fundamentally these machine AIs are all doing the same thing regardless of their precise methods: performing math really really quickly and efficiently, and their performance is ultimately going to depend on just how difficult and complex the underlying math of a given game is. Also ultimately it doesn't matter that much because playing games is extremely easy insofar as machine-learned behaviors go because all games are little more than giant obfuscated math problems, but that's a different topic altogether.
|
# ? Nov 25, 2019 02:58 |
|
Super Jay Mann posted:Fundamentally these machine AIs are all doing the same thing regardless of their precise methods: performing math really really quickly and efficiently This is a reductive and absolutely unhelpful way to describe how a neural net works. It'd be like saying "all your brain does when it plays games is a bunch of chemical reactions and electricity". Super Jay Mann posted:their performance is ultimately going to depend on just how difficult and complex the underlying math of a given game is. Neural nets don't play games by understanding the underlying math of a game any more than you understand games that way. Games like Nim have extremely simple underlying math, but can still be hard for neural nets to figure out because the NN only sees positions, how they connect, and whether they win. The underlying math of the game is a poor predictor for how easy it will be to approach through a NN type approach. Super Jay Mann posted:Also ultimately it doesn't matter that much because playing games is extremely easy insofar as machine-learned behaviors go because all games are little more than giant obfuscated math problems, but that's a different topic altogether. Games are not extremely easy for AIs. It has taken 50 years of hardware and software development to get to the point we're at now, where complex games can be beat without game-specific, human-designed approaches. It is still beyond almost anyone in the world to create these kinds of successful agents for the kinds of games humans find interesting. As machine learning technology becomes more robust and accessible, it will find its way into a tremendous amount of real world decision making.
|
# ? Nov 25, 2019 05:20 |
|
Super Jay Mann posted:Fundamentally these machine AIs are all doing the same thing regardless of their precise methods: performing math really really quickly and efficiently, and their performance is ultimately going to depend on just how difficult and complex the underlying math of a given game is. This is reductionist. "All biology is really chemistry, and all chemistry is really physics, which all extends from the properties of fundamental particles, so why really bother studying anything beyond the fundamental particles." It's because a) not even the most advanced machine from 50 years in the future will be able to simulate an entire bear from fundamental particles and b) higher order effects also follow patterns that are maybe less rigid, but can still be usefully studied. The question is less "can the AI solve the problem" (which it probably can't do that well (see the relatively simple chess), but that's for later), and more "how well can the AI get through the layers of obfuscation".
|
# ? Nov 25, 2019 05:25 |
|
jmzero posted:This is a reductive and absolutely unhelpful way to describe how a neural net works. It'd be like saying "all your brain does when it plays games is a bunch of chemical reactions and electricity". I'll concede this point, but only insofar as NN approaches are just substituting one mathematical approach (games modeled as functions) with another (games modeled as discrete, interconnected positions which can be clustered into patterns and "learned"). And that's great! There's great value in developing an approach that abstracts the process of determining a winning series of moves in relatively quick fashion regardless of the actual mechanics of the problem in question, but I don't think that changes my general point that the efficacy of the algorithm is still dependent on the complexity of the problem (just what kind of complexity we're talking about changes depending on the approach) and that games are quite easy compared to many other real-world problems that have been modeled. quote:Games are not extremely easy for AIs. It has taken 50 years of hardware and software development to get to the point we're at now, where complex games can be beat without game-specific, human-designed approaches. It is still beyond almost anyone in the world to create these kinds of successful agents for the kinds of games humans find interesting. As machine learning technology becomes more robust and accessible, it will find its way into a tremendous amount of real world decision making. To clarify here, easy does not mean fast. Many games are structured with players have perfect and/or complete information, so they can easily be solved simply by analyzing the results of every single possible move from every single possible position. The computation time of doing even for simple games is usually so astronomically high so as to be nonviable with anything but the best computers, but that doesn't change the actual difficulty of the problem. Most real-world problems don't have the luxury of being so straightforward. Which is why the major advancements in AI usage for real world scenarios in the past several decades was/isn't just about improving the algorithms, it was/is about learning how to reframe these real world scenarios into simplified models that our dumb computers could actually work with so they can provide answers that are "not quite rigorous but close enough". I suppose thinking about it now, that makes games a good stepping stone for AIs since it appears humanity's approach to these difficult real world problems to better manage them is to "gamify" them. Dancer posted:The question is less "can the AI solve the problem" (which it probably can't do that well (see the relatively simple chess), but that's for later), and more "how well can the AI get through the layers of obfuscation". This but also add to that "How can we make the problem itself easier for the AI to work with?" The computer can't do it alone, we need to help them. I'm sorry for the aside though, it's an interesting topic to me but I'm not sure how much it fits in this thread specifically.
|
# ? Nov 25, 2019 06:55 |
|
Bottom Liner posted:If anyone sees Food Chain Magnate Ketchup in stock online (CSI/MM/Gamenerdz) please post here and give me a heads up. Cardhaus has it for $125. They are reputable, I've ordered several things from them. Also, linking my BGG auction here. If you see something that you want I'll cut you a deal below the BIN, assuming you live in the US. At the very least, drop me a thumbs up for visibility. Shadow225 fucked around with this message at 07:37 on Nov 25, 2019 |
# ? Nov 25, 2019 07:10 |
|
People who say sorry for the aside never sound very sorry
|
# ? Nov 25, 2019 07:57 |
|
Shadow225 posted:Cardhaus has it for $125. They are reputable, I've ordered several things from them.
|
# ? Nov 25, 2019 08:06 |
|
|
# ? May 11, 2024 15:51 |
|
Super Jay Mann posted:Fundamentally these machine AIs are all doing the same thing regardless of their precise methods: performing math really really quickly and efficiently, and their performance is ultimately going to depend on just how difficult and complex the underlying math of a given game is. The big problem for AI's in general is they do not know what they are doing, they have goals, maybe it's scoring points, maybe it's not losing the game, but fundamentally it does not know it's playing a game or the context for it. That can cause it to do really weird things, like the Age of Empires AI that would intentionally crash the game because that was not a 'lose' for it.
|
# ? Nov 25, 2019 10:16 |