scr0llbaer

It should work when you give it the original as input (attached file)? Today's "AI" are not search engines although people misunderstand them as all-knowing oracles (and then complain about the hallucinations). Commercial AI providers of course try to accommodate for their customers and fulfill their expectations, but this prevents casual users from developing a good intuition what this tech actually is and can do. Neural-network-AI works much like the human brain, it went through training but it forgets or overlooks things which are not in immediate focus. The training data is not some internal database, the "training" just "shapes" the "neurons". It may "remember" sports results of popular matches correctly because often mentioned in many sources, but not the obscure ones, and may hallucinate them instead of realizing it just doesn't have the information, which is another problem the tech had for a long time, "AI" doesn't know what it doesn't know. This has been somewhat solved by know when they ask for external sources (web search) (even local models nowadays do web search if they feel they don't know something), but this is still not 100% reliable. Long story short, always give these things as much input as possible yourself (even if it means a lot of work that is often promised, falsely, that "AI" can "replace").

It's the Comic Sans age of "AI", these are all explicit (presented by the AI provider as such) or implicit (from the model training) presets. One reason is the provider thinks some presets create fewer errors than others. It's funny how each generation of e.g. the OpenAI/ChatGPT image tool produces its own distinct look that has nothing to do with the one from the previous iteration. If you try to prompt a different style it just looks more like a filter applied. Things have been slowly getting better in that regard though.

Ah yeah this map, I guess it has to mercator-project even harder for proper balancing.

lol wasn't this thread originally just about some map editor thing with a provocative generalized title? now this thread was "hijacked" to match that title. still, maybe these topics could be separated. Fast clicking is not the same as building out fast btw. Even in old turn-based games like Civ1 you're gonna lose if you "idle" (though admittedly you'll be much more aware of "idling") and you learn to optimize your build-out-strategy. So in principle there is no way around learning to build fast enough in this type of game. Also for the AoE series I see a lot about "booming" doing web searches. Maybe there could be different game modes, but these will require different balancing. 1vs1 is already quite different than 4vs4. I like to play against many (allied) AIs on a giant map because I just like the epic-ness of that, the "history" and tales that may unfold in your head here. The game is not very balanced or fine-tuned for that but I can "improvise". Finally you get to meaningfully build walls and all for better protection of your bases while most of your army is campaigning away, that sort of thing. But maybe this is the experience some players are looking for and what the original documents (things should never be set in stone btw anyway) hint at when referring to "historic accuracy" and things like that. Problem of course, it's an open source project, and for general gamers not familiar with that, this sort of "free games" have a reputation of not being very polished, and it's true in the case of 0 A.D., the staff is lacking to cater properly to newcomers, and there isn't authoritative quality control. One of my opened issues was closed (perhaps I should have discussed in the forums first admittedly) even though it is a CLEAR AND OBVIOUS thing for anyone even remotely interested in good game design. But yeah, "typically open source".

Yes, "AI" has been a term since the 1950s at least. The grand vision of an "AGI", an Artificial General Intelligence, basically the SciFi scenario of an all-knowing all-capable entity, already existed back then. Many developments in computing that were thought to help achieve this vision hence ran under an "AI" label historically, not only neural networks, the stuff that LLMs and picture generators of today are based on today and that eat so many resources. Chatbots like ELIZA were not based on neural networks. Just classical IF-THEN-ELSE and some clever play on human psychology. But yes, the neural network approach was also known since at least the 1950s, but it was all theory, it was known the required computing power would be infeasible for a very long time. Today's web and Google didn't exist back then, so it was assumed a more traditional knowledge base expert system would also be a necessary component of an AGI, hence many developments apart from resource-eating neural networks also ran under the "AI" label, but I'm almost certain the term AI in video games didn't even stem from that, i.e. from research lab level algorithmic sophistication, it was rather, as said, just a colloquial way to say "non human opponent".

"AI that makes you laugh", "AI that has personality", haha, these are all quite moonshot things. Not even many commercial AAA games are that sophisticated. Are people so used to what "AI" can supposedly do nowadays that demands are so high? It's just a terminology coincidence of course, these are not LLMs, computer opponents have been called "AI" throughout practically all of gaming history, "artificial" just meaning "non-human" here, to differentiate from player vs player modes like in Street Fighter or the like. LLM functionality in a game like this would not be practical, they're expensive and someone has to pay for the tokens. On the local machine it would use a lot of performance and even if workable, it will be very hard to design it for the performance levels of different machines users have. But I have seen a thread with a mod experimenting with this. (And of course it will be two layers, grunt work still done by Petra, LLM deciding on the grand strategy "in the background"). Sure, the "AI" in the game could be improved, but it still lacks very basic things that are more important, like spatial awareness, looking at the map and figuring out what the choke points are. But these are not trivial problems to solve. Maybe LLMs can help with some "vibe-coding" to improve the current state but that's it.

It's already on itch.io at least.

It might be a viable strategy and just a way of preference and for sure it's more than sufficient against just one AI opponent, but in other settings it might be important to get certain upgrades (market, forge,...) sooner.

Icon styles in computing history went from simple to rich in gradients and effects (Windows XP and the like, "because we can") back to simple, but now with a lot of perception psychology knowledge put into them. Most 0 A.D. icons are very 1990s, even 1980s-ish.

Hello, yes we've all been there! For new players, this game inspires some casual ancient city building, but it's really about building out fast in the beginning. There is some discussion in this forum about how to make the game more accessible, it may be just about terminology (renaming the difficulty levels), though I guess modern gamers are also used to a rubberbanding AI that adapts to the player's gameplay, which is what bigger commercial games do. "No idling units! Place things well to keep distances short, running around is not productive! The first barracks need to be ready at minute 4! Your civ center and your barracks need to churn out non-stop! (Use auto-queue feature.) 100 pop (about 50 civilians, 50 citizen soldiers) at minute 10 or earlier! Then you'll have no problem with the AI opponent. (Phase 2 at around minute 8, second barracks and third barracks or stables before minute 10.)" This is the gist, in my opinion. Maybe info like this should be in the map selection screen. The tutorial, if paid attention to at all, also doesn't get that message across. It's not good "game design" if this info has to be researched on the web. So that means early game is balancing working food, wood, and building houses fast enough. When the auto-queue breaks or you have hit the housing limit you'll know there is room for optimization. There is an option to enable in-game time to be shown on the top right, this should be enabled by default in my opinion, because it is central to this game. You'll find each civ is a bit different, but not much. Romans however aren't exactly the easiest civ because they have swordsmen as the only melee unit in the beginning, and they need iron, so you need to manage that early. Usually (with other civs) you don't have to start mining stone and iron before minute 7. A few need early stone for the slingers though. (Also Romans lack proper ranged units).

When I started playing the game I also wondered about the food icon and that it might put off some people, so, as expected. Personally I'm not vegan, but I also think, for the standard game, a wheat icon would be more representative, because arguably, most food in most cases comes from working the fields. And it's pretty standard in other games too (Civ etc). Now, distinctive colorization was mentioned somewhere as an argument against the wheat symbol. Not so convincing to me because the wood's brown is quite in the middle between yellow and red in color space. Also there are a lot of similar problems in 0 A.D.'s iconography. Docks, Barracks, Storehouse icons are hard to distinguish, color-wise. I still need to read tooltip distinguishing FIeld from Farmstead etc

More like, zombies? Speaking of which, a zombies mod could be nice.

The "voice acting" in 0 A.D. indeed could see some improvement. The quality is all over the place. Some is "good enough" that it doesn't bother, but with some voice samples you really hear it was recorded with a bad mic in a closet with the voice muffled and dampened. Some of the voice "acting" really funny where it should be a call to arms or battle cry but it sounds like they're reading it from a book.

Sure, just commenting on a technicality that might be a common misconception or just a "never thought about that"-thing. Who cares about those "large software houses", they were ruining gaming. I don't follow and I don't care about modern gaming much, I'm into retro and open source. About this "AI decomp revolution" in general: Both a curse and a blessing. Back then, music could suddenly be produced in the bedroom, rather than just in expensive studios. So, lots of "slop" in many people's eyes (or ears) back then too. But you don’t have to - and you don’t need to - consume everything that comes out. Even as an experienced programmer, you often don’t have the time or the inspiration to indulge in developing some hobbyist retro projects in your free time. That’s easier now. Projects often lie dormant for years because you’ve hit a roadblock somewhere or would have to put in a lot of work to overcome a certain technical hurdle, so you end up procrastinating. This can now be largely overcome "thanks" to "AI". On the other hand, of course, it also opens the floodgates to endless slop and half-finished, unpolished stuff which initially is touted with great enthusiasm. Development is 20%, maintenance / quality assurance / project management is 80% - and many “vibe coders” either lack the motivation, the skills, the experience, or all of that. One solution might be curation - regularly showcasing good projects that are truly worth checking out. Many retro sites have always tried to cover everything that comes out, but that might not work anymore. What I’d like to see is a specific project - a piece of software or an app, similar to a launcher - that contains a library of consistently well-curated picks of “Decomps” and the like.

Programming languages are for humans. Computers already do machine code, so no need to "decompile" to understand. "AI" doesn't need human-readable programming languages either. Although LLMs are based on neural networks ("how human brain works"), they nowadays probably were trained on enough native machine code as well. So what "AI" nowadays does is providing more "bridges" for humans.