|
What is supposed to be bad about that timelapse?
|
#
?
Jun 4, 2016 12:49
|
|
|
|
| # ? Jun 7, 2024 17:59 |
|
|
The bits where a (single?) division pushes half way through Europe and the AI doesn't cut it off or seem to respond to it for 6 months?
|
|
#
?
Jun 4, 2016 12:55
|
|
|
While I'm glad they localized country names based on government type, there are a lot that look pretty ridiculous. "Empire of the Platypus" = Fascist Australia "Union of Red Dragons" = Communist Bhutan Really want to know if Canada is like the People's Republic of the Beaver or some dumbass poo poo.
|
|
#
?
Jun 4, 2016 13:01
|
|
|
Drone posted:While I'm glad they localized country names based on government type, there are a lot that look pretty ridiculous. This is actually amazing and better than any normal-rear end localization. EDIT: Red Beaver Commune
|
|
#
?
Jun 4, 2016 13:03
|
|
|
Wooper posted:What is supposed to be bad about that timelapse? The German AI couldn't finish off the Turkey AI in seven years of grueling warfare. How do you think it will fare against a naked primate with an oversized brain and a basic understanding of tactics?
|
|
#
?
Jun 4, 2016 13:23
|
|
|
waitwhatno posted:The German AI couldn't finish off the Turkey AI in seven years of grueling warfare. How do you think it will fare against a naked primate with an oversized brain and a basic understanding of tactics? Tactics aren't even necessary. You can eliminate Germany with Poland before 1940 with nothing but infantry and a wide advance from the eastern border to the west.
|
|
#
?
Jun 4, 2016 13:31
|
|
|
Funky Valentine posted:EDIT: Red Beaver Commune Okay I'm modding this in. As far as the AI goes, I've seen a few people who recommend that those with previous Paradox game experience play on the harder difficulty. That's not encouraging. Edit: okay so I've also heard from streamers that the release build has gotten some kind of AI overhaul compared to the pre-release one. Drone fucked around with this message at 13:45 on Jun 4, 2016 |
|
#
?
Jun 4, 2016 13:38
|
|
|
Darkrenown posted:I dunno why you guys are mad about the highly advanced AI which accurately simulates the historical hilarious levels of incompetence and ill-preparation found in every WWII combatant. Abandoning fortified lines or leaving fronts almost or entirely unguarded doesn't stand out as any weirder as the poo poo that actually happened. Strategic cockups in WWII were more in the vein of "We can totally take this town 30 miles behind the front with airborne forces and then move in with tanks, the Germans are finished, war will be over by Christmas" than "WE MUST DEFEND THE SWISS BORDER WITH EVERYTHING WE HAVE, NEVERMIND THE OTHER ENEMIES WE HAVE!"
|
|
#
?
Jun 4, 2016 13:58
|
|
|
Drone posted:While I'm glad they localized country names based on government type, there are a lot that look pretty ridiculous. I just saw DDRJake's Union of Red Dragons occupy Afghanistan and I have no complaints. Graveyard of empires, pfah. Also, what should the Land of the Thunder Dragon be named after the communist revolution?
|
|
#
?
Jun 4, 2016 14:17
|
|
|
ArchangeI posted:Strategic cockups in WWII were more in the vein of "We can totally take this town 30 miles behind the front with airborne forces and then move in with tanks, the Germans are finished, war will be over by Christmas" than "WE MUST DEFEND THE SWISS BORDER WITH EVERYTHING WE HAVE, NEVERMIND THE OTHER ENEMIES WE HAVE!" "Let's give Hitler this land he wants, he promises that's the last bit he'll take!" "Massive German build up on my border and everyone keeps telling me they are about to invade? Nah, they are our friends!" "Let's push well past our fortified river line in Belgium! Also, what kind of loser needs a strategic reserve? Also also, gently caress that flank, no way tanks can go through a forest" "Let's declare war on the USA, they are weak!" *2 "I bet the USA will surrender if we sink their battleships!" "Let me just stand down the army but then suddenly declare war on Greece and order an invasion!" "Let's just put the whole army into Albania to counter-attack Italy while Germany sweeps through Yugoslavia and Bulgaria joins the Axis!" "It's not like Britain will go to war over Poland!" "Britain will certainly surrender now that France has been defeated!" "We only have to kick the door in and the whole rotten structure will come crashing down!" "We should fight house to house for this city, bleeding our army white while our flanks are guarded only by our severely overextended, badly trained, and unequipped allies. WCGW?" "Germany has basically won the war, we better join in so we can get something in the peace deal!" "Our Libyan army can totally take Egypt and bring back our empire!"
|
|
#
?
Jun 4, 2016 15:14
|
|
|
Most of those are geostrategic mistakes, not mistakes made by what would be the unit deployment AI in game. I haven't seen the AI make geostrategic mistakes yet, so good work on that. I'll give you Italy, but I'm not sure if "Our AI is as bad at actual warfare as the most incompetent country in WWII" is a particularly good selling point.
|
|
#
?
Jun 4, 2016 15:45
|
|
|
Just finished watching all of Quill18's videos of his Italy run so far. Towards the end he was even talking about how the AI was lackluster. At one point, Hungary was at war with Germany, and the German AI did basically nothing to stop them for the first few months of the war. I dunno man. I'm super optimistic but it's going to be one of those things that I won't be able to properly judge until I'm playing the game.
|
|
#
?
Jun 4, 2016 15:47
|
|
Drone posted:Just finished watching all of Quill18's videos of his Italy run so far. Towards the end he was even talking about how the AI was lackluster. At one point, Hungary was at war with Germany, and the German AI did basically nothing to stop them for the first few months of the war. Actually I think that Germany in his playthrough is not really at fault for the sorry state the Axis is in but he himself is. By taking a large chunk of France he catapulted himself way ahead of Germany in terms of industrial strength but is not supporting his allies in subsequent wars (e.g. against Hungary) but rather uses his strength to conquer all of Africa (Why?), Turkey and parts of the Middle East. Germany having troubles with Weserübung is a bit disconcerting but without a look at a composition of the German forces and where they are needed, you can't really blame the AI too much, especially as it's an AI vs. AI situation, which means that Hungary's and Denmark's AI performing very good.
|
|
|
#
?
Jun 4, 2016 16:02
|
|
|
Darkrenown posted:"Let's give Hitler this land he wants, he promises that's the last bit he'll take!" Darkrenown, we all love you and so on, but nobody wants to play a game against a wacky and random AI that acts like it graduated from the GBS school of pants-making GBS threads tactics. That's not fun.
|
|
#
?
Jun 4, 2016 16:02
|
|
|
Who playing day 1 multiplayer?
|
|
#
?
Jun 4, 2016 16:10
|
|
|
Pop and Loch Nessy posted:Who playing day 1 multiplayer? Sounds good
|
|
#
?
Jun 4, 2016 16:15
|
|
|
I've been watching a few streams and I never realized how bad the average person is. Like this guy who's trying to conquer China as Japan with only 24 regiments because he needs the other 50 to garrison random islands in the Pacific even though he's not at war with America and likely won't be for years. No naval invasions on China's huge coast. Doesn't seem to realize simply camping out in Beijing without pushing any farther south will never end the war. Like is constantly checking to see if he can negotiate peace and doesn't understand why China is telling him to gently caress off.
|
|
#
?
Jun 4, 2016 16:15
|
|
|
did they remove practical knowledge in research?
|
|
#
?
Jun 4, 2016 16:18
|
|
Descar posted:did they remove practical knowledge in research? Yes. But production efficiency is a similar concept.
|
|
|
#
?
Jun 4, 2016 16:22
|
|
|
ArchangeI posted:I'll give you Italy, but I'm not sure if "Our AI is as bad at actual warfare as the most incompetent country in WWII" is a particularly good selling point. waitwhatno posted:Darkrenown, we all love you and so on, but nobody wants to play a game against a wacky and random AI that acts like it graduated from the GBS school of pants-making GBS threads tactics. That's not fun. I know! Remember how I said: (No I'm not entirely serious, obviously you want the AI to be decently challenging to be fun, but massive strategic cockups are part of the WWII parcel) Just making bad jokes about games, I don't actually want the AI to ignore fronts or anything.
|
|
#
?
Jun 4, 2016 16:31
|
|
|
Drone posted:While I'm glad they localized country names based on government type, there are a lot that look pretty ridiculous. No, it is just "People's Republic of Canada" An Youtuber is doing a Communist Canada play through.
|
|
#
?
Jun 4, 2016 17:04
|
|
|
Does Communist Canada get kicked out of the Allies?
|
|
#
?
Jun 4, 2016 17:28
|
|
|
See it's a feature... I think my go to tactic will be building a billion infantry and one light armor. The infantry will make a huge front and the light armor will go through the spots the AI forgot to defend.
|
|
#
?
Jun 4, 2016 17:44
|
|
|
EU4 is possibly the only game I've played where I actually liked going for achievements. Is the list on the HoI4 wiki all of them in this one? Seems a bit meagre, but at least
|
|
#
?
Jun 4, 2016 17:44
|
|
|
I'm 100% going for 'Duce Nuked'em' first
|
|
#
?
Jun 4, 2016 17:47
|
|
|
Xerophyte posted:EU4 is possibly the only game I've played where I actually liked going for achievements. Is the list on the HoI4 wiki all of them in this one? Seems a bit meagre, but at least The added a lot more EU4 achievements since release so I'm sure we'll get more.
|
|
#
?
Jun 4, 2016 18:00
|
|
|
There's probably going to be plenty of psuedo WC achievements like own all the world's oil as madagascar or something
|
|
#
?
Jun 4, 2016 18:08
|
|
|
I like picking Italy and staying out of the war. I just go around taking over the Balkans and puppet Turkey (Dardanelles are mine though) and generally just watch the chaos unfold.
|
|
#
?
Jun 4, 2016 18:15
|
|
|
Darkrenown posted:I know! Remember how I said: Maybe I'm too jaded and can't see it as anything but a developer going "nono, you don't understand, we had an AI that was so awesome it beat everybody so we had to artificially dumb it down and force it to make mistakes", which usually means their AI sucks and they are scrambling for an explanation.
|
|
#
?
Jun 4, 2016 18:28
|
|
|
Xerophyte posted:EU4 is possibly the only game I've played where I actually liked going for achievements. Is the list on the HoI4 wiki all of them in this one? Seems a bit meagre, but at least You can get Wojtek as a commander 
|
|
#
?
Jun 4, 2016 18:30
|
|
|
I expect a little AI wonkiness if I put a skill 1 general in charge of an entire theater; couldn't skill 1 generals have special bad general events? Not necessarily https://en.wikipedia.org/wiki/Pietro_Maletti bad, but maybe https://en.wikipedia.org/wiki/Lloyd_Fredendall bad? Ghetto Prince fucked around with this message at 23:13 on Jun 4, 2016 |
|
#
?
Jun 4, 2016 18:34
|
|
|
Xerophyte posted:EU4 is possibly the only game I've played where I actually liked going for achievements. Is the list on the HoI4 wiki all of them in this one? Seems a bit meagre, but at least And the icon for the Swedish achievement is the Barsebäck nuclear power plant that can be seen from Denmark.
|
|
#
?
Jun 4, 2016 18:36
|
|
|
Bundle of Keys posted:And the icon for the Swedish achievement is the Barsebäck nuclear power plant that can be seen from Denmark. Yes, it features in the scene from Riget that's being referenced. I glanced at the collapse of the Reich at DDRJake's Bhutan stream and the AI really couldn't handle that. Too many pockets and too many borders with too many nations meant everyone's divisions mostly just shuffled around trying to find a province to guard. Notably, Germany left Berlin open.
|
|
#
?
Jun 4, 2016 18:49
|
|
|
After watching a bunch of streams I'm convinced that the issue is that the AI doesn't create enough armies/fronts and then gets overwhelmed and has no clue what to do. The AI works well until it gets into situations where it has too much frontage and it wigs out.
|
|
#
?
Jun 4, 2016 18:56
|
|
|
Actually, I had a thought. All the time lapses have been centered on Europe, which is a given that's where all the poo poo goes down for the most part. Europe is all one body, however, and while the US does notably manage to invade Europe in some fashion (Even via Ukraine in that one, which is super loving impressive imo) it's all one giant mass of land. Is there a time lapse for the Pacific? How does the AI handle all dem fukken islands? For that matter, seeing Japan and China duke it out might be a good way to see how the AI actually handles as opposed to Germany going in like three loving directions at once, which is a given to mess up somewhere.
|
|
#
?
Jun 4, 2016 19:58
|
|
|
Pop and Loch Nessy posted:Who playing day 1 multiplayer? Sure why not.
|
|
#
?
Jun 4, 2016 21:15
|
|
|
Stairmaster posted:Sure why not. I'm in dude
|
|
#
?
Jun 4, 2016 21:17
|
|
|
Okay!
|
|
#
?
Jun 4, 2016 21:24
|
|
|
I call Mongolia. Nothing but horse cavalry regiments.
|
|
#
?
Jun 4, 2016 22:07
|
|
|
|
| # ? Jun 7, 2024 17:59 |
|
|
Slaughterhouse-Ive posted:Maybe you can't comment on this but is AI a second-class citizen when designing these types of games? I don't mean that in a bad way. I feel like there's a pretty heavy config-driven element to AI development and you can't have AI without the mechanics in place so occasionally goofy AI behaviors don't particularly bother me as much if the mechanics and core gameplay are solid. My current work involves AI development but not games related so I love sperging out over this sort of stuff. I can say that games take a very different approach in general due to their different goals and working under different constraints. Take a look at the earliest games and you'll quickly notice that the enemies all move in fixed patterns. They might shoot bullets straight ahead or in some other sort of pattern, but they do it like clockwork. A tiny bit more sophisticated enemies add a bit of randomness, like by once in a while jumping. The randomness doesn't make them smarter, but it looks less fake. Enemies in a game like Final Fantasy just pick a random ability off a list every turn and pick a random target. Many video game bosses randomly switch between patterns. What all these have in common is that they take almost no processing power at all per frame. Games got slightly more sophisticated by adding capabilities like enemies aiming at the player or moving towards the player. There's no pathfinding yet so they'll usually just run into walls, but it's something. Another innovation is enemies having a default state, like resting or walking in a pattern and then changing the pattern to something else when the player gets too close, like chasing them. While this may sound primitive, it's actually still very widely used because it's computationally cheap and can be very convincing by combining it with more sophisticated techniques for specific applications. So play a single-player FPS, stealth shooter or whatever, and you'll notice that the enemies do various stuff when they don't notice you. That's probably a simple script running as part of their resting state. If they notice the player or something else, they might go into a searching state with the location of the disturbance as a target. They might also trigger the same state in nearby enemies as well. If they don't find anything after a timeout they return to the resting state or if they spot the player in that time they go to the fighting state. The fighting state may have multiple substates that they transition between depending on the circumstances. So if they're in the Under Cover state and the player throws a grenade, they might go into the Dodge Grenade state. Otherwise they might shoot from cover. There are actual AI techniques involved like if the player starts to run away they'll use pathfinding to follow, but special purpose heuristics often work just as well while being faster and easier to implement. They're working in a very constrained design space, and can use various techniques to further simplify things. Like the whole idea of being able to snap to cover is much easier to work with than trying to analyze the local level geometry to tell if the player's hiding behind a wall. Games have started moving to more sophisticated techniques like Goal Oriented Action Planning (GOAP) to start giving something more resembling genuine smarts but uptake isn't instantaneous. Planning differs from previous techniques by guessing the results of possible actions and looking ahead to find an efficient combination to reach a given goal. Generating a plan is expensive but it's cheap to just follow a plan. But the big thing is that it allows some level of foresight not possible with other techniques. For a game like Hearts of Iron, AI is actually a very high priority, but the problem is that it's much harder to make convincing because it has to actually be good at some level and that's a different story. F.E.A.R. was better than previous games, but it also took advantage of techniques like soldiers shouting out things based on their goals and plans to show the player that it's happening. For the sake of performance, strategy games still need to rely on special purpose heuristics and similar techniques to take the massive amount of information to be processed and reduce it to something traditional AI can handle in a reasonable amount of time. So it's not a second-class citizen, just difficult. There are techniques that could be used to make the AI more convincing even if it's not necessarily better, but would potentially require significant additional processing power. This next bit isn't meant to be critical of the game or saying I would do it better, just effort posting about a topic I find really interesting and maybe helping people understand AI better. The paper I posted about GOAP also talked about the distinct squad level AI in F.E.A.R. The squad level AI would generate goals for each enemy within the squad which would then generate plans in response. Then it would monitor their progress along those plans by comparing the expected world state to the actual world state. The paper also notes that the squad level AI was pretty ad hoc but the same GOAP architecture could be applied to it as well so that there are squad level goals and plans. This can be further generalized into something called a hierarchical task network, or HTN. HTNs take a high level goal and recursively split it into plans consisting of lower level goals that are assigned to the agents the next level down. This process is repeated until it reaches the bottom level agents which make plans consisting of actions that they can perform. This means that there's a lot more goals to generate plans for, but each agent is working in a much more constrained world model. Each agent also monitors its child agents and the relevant world state to see if it reflects the plan, and generate a possible reason if it doesn't. If it generates a new plan, the reason makes it prefer actions that should be more successful. HTNs also map very well to military chains of command where each agent is a unit and its children are subordinate units. Only considering 3-5 subordinate units drastically reduces the search space for each unit compared to a unified AI trying to consider every unit at once. The commanding officers can provide a face for the AI, explaining why it made specific decisions and giving the player the ability to reprioritize goals. This could be accessed by a button labeled WHAT THE gently caress WERE YOU THINKING!? Soviet players would of course have additional catharsis-providing options available.  How it might play out in practice is that the player or national AI might have a goal of "Defend against possible German invasion" which is passed to the theater commanders. Each theater commander looks at their territory and creates goals based on that like "Defend the German border" or "Defend major cities" which it hands off to subordinates. If some unit isn't able to create a plan, like if it doesn't have enough troops to guard the border, it can notify the unit above it which revises its plans in turn. Those revised plans might reassign goals or reshuffle troops based on the relative strengths. If those fail, it continues its way up until someone can do it or it reaches the player, telling them that they need more troops for instance. I guess I could go into more detail but this is already pretty drat long.
|
|
#
?
Jun 4, 2016 23:01
|
|
