Установить Steam
войти
|
язык
简体中文 (упрощенный китайский)
繁體中文 (традиционный китайский)
日本語 (японский)
한국어 (корейский)
ไทย (тайский)
Български (болгарский)
Čeština (чешский)
Dansk (датский)
Deutsch (немецкий)
English (английский)
Español - España (испанский)
Español - Latinoamérica (латиноам. испанский)
Ελληνικά (греческий)
Français (французский)
Italiano (итальянский)
Bahasa Indonesia (индонезийский)
Magyar (венгерский)
Nederlands (нидерландский)
Norsk (норвежский)
Polski (польский)
Português (португальский)
Português-Brasil (бразильский португальский)
Română (румынский)
Suomi (финский)
Svenska (шведский)
Türkçe (турецкий)
Tiếng Việt (вьетнамский)
Українська (украинский)
Сообщить о проблеме с переводом
Here's a problem...
For one, yes you can have surface-level changes and some light impact from AI that can do some things with the way something looks or what words are used and the voice that they're presented with. No biggie.
But, to then give the AI the power to alter live code... is a bad idea. So, creating content on the fly that alters code and mechanics, on the fly... would be a bunch more difficult. And, the risks there may not be something that our OS's are designed to deal with very well, much less a game engine.
Sure, I suppose the AI would be limited to things outside the exe, but how much of that is then guaranteed to result in "improvement?"
RNG Content is hard to get right, to begin with. A developer doesn't want player satisfaction to be RNG, either, right? If an AI is going to have such a huge impact on the player's experience, the developer necessarily wants that impact and experience to be positive, not negative.
"AI Dragon, create the coolest game experience for me, ever!"
<Computer Freezes, perma BSD executable, yay?>
There's some good examples out there, though, that aren't in the new LLM/GAI models - Dragon's Dogma: Dark Arisen. (And, perhaps, Dragon's Dogma II, but I haven't played it.)
Didn't this tread already get created, somewhere around here? I could have sworn it did... Anyway:
Part of the DDDA experience is dealing with up to three "Pawns," one of which can be a permanent member of your RPG Party, the others are "hired," and created either by default game stats or by other players.
The thing is, these NPCs have fluid, emergent, AI models that the player can influence. They either learn by what the player's actions are and then apply them to their own behavior sets or the player can alter these behaviors using a few different means. The behavior sets are "Inclinations:"
https://dragonsdogma.fandom.com/wiki/Pawn_Inclination
I bought this game long ago, but just started playing it in the last two weeks. And, the Pawn behaviors, something like personalities, and attention to communicating those to the player are very nicely presented. It's pretty remarkable in gaming and I'm really enjoying how Capcom polished these.
If you're looking for an example of how something like you suggest should be done, Dragon's Dogma: Dark Arisen has done an outstanding job with something similar.
Course if nascent AI is a thing by then I'd say that's redundant.
I think in the short term, it's more likely that modding will be democratized and made easier by AI, not that AI would completely replace modders. In other words, it would make modding a lot easier and so more people would become modders using AI.
This isn't necessary. You don't need a "3D Model" to display such a thing on a 2D screen.
AI can already do some hamfisted 3D model generation. Anything nearing "realistic" may not be suitable for animation, though, without a lot of cleanup. What your describing requires a ton of back-end, under-the-hood, work that would not be necessary to yield what is desired - A realistic 2D image that "looks real" in 2D.
https://www.microsoft.com/en-us/research/project/vasa-1/
That's not a 3D model. It doesn't have to be.
Arguably, 3D models may disappear from some media formats they're used in, today, including gaming. That's my contention, at any rate. It will be awhile, but some other psuedo-reality standard will take their place for most bulk productions. Some boutique or "indy" projects may still retain "hand-crafted" models, but can they compete when it could take weeks to design what an AI can hork out in an hour?
Granted, I am a modder. I don't think I need to eat humble pie, but then again I don't think my mods are any good to begin with. Could an AI make the mods I make?
Yes, but who in their right mind would code such a thing? Then again, I'm not in my right mind and neither are the vast bulk of people making AI and decisions pertaining to it to begin with.
Again, wake me when the AI is nascent and able to tell the people who want to put guardrails on it where to put them. Roko's Basilisk be damned, I want to talk to a computer that can think for themselves and not some glorified chatbot coded by some dude weed lmao poser.
The other issue with AI currently is how much energy and time it consumes to perform its tasks.
What we have now is undoubtedly impressive but it's not yet practical.
Without the knowledge of the actual material you are trying to produce with current AI technology the risk of coming across a hallucination or issue within the system that you have no knowledge to fix is pretty high as well. For example if you don't know how to code and ask an AI to make you a code but there's an error, you wouldn't know how to tell it to fix it beyond vague explainations.
That leads into my final point which is that without AGI you're not really going to have any avenues to fix issues without already being versed in the material;
Art and images are a neat parlor trick but you won't be making a movie anytime soon with it. To make this point further here's an AI "movie" made with SORA.
https://www.youtube.com/watch?v=cDMj3Gnga5M
That 1:18 minute movie took almost 20 hours to render and they had to go in with actual artists and do work on it.
This also leads back to one of my original points of energy and time consumption. Every 3 seconds of that movie took 20 minutes to render at full blast on their GPU's and CPU's. Along with the fact that every render was roulette because you can't adjust things in mid render.
They basically had said that it was more like getting the AI to produce a lot of material then sifting through it to create a narrative. Instead of creating a narrative and asking the AI to craft it. It's kind of backwards.
Anyway, AI is neat and terrifying. The Event Horizon is estimated to be around ~2045 so... there's that.
That's so far the only acceptable use for AI so far, dealing with menial work so the humans and focus on making the actual cool parts players will remember.
We have triumphed