Posted: 1/23/2026 11:18:01 PM EDT
|
Take a classic like Skyrim or any other classic RPG. I think in the near future, AI will make NPCs so much more interactive. Instead of a handful of terse, limited responses from essentially a walking piece of furniture, every "person" you encounter in a game will be sble to carry on a conversation. I also think you will be able to upload pictures of yourself and the games will be able to generate an accurate facsimile so you can play as "yourself" instead of tweaking facial features. What other changes will we see as AI tech is applied to gaming? |
|
Originally Posted By Hayashi_Killian: Wish in one hand. Originally Posted By Hayashi_Killian: Originally Posted By Jambalaya: I think in the near future, AI will make NPCs so much more interactive. Wish in one hand. Unfortunately, I think you're dead wrong here. Not that it wouldn't make gaming awesome, but AI at that level is going to touch a lot more than gaming (obviously), and in a very bad way... I absolutely believe GTA6 got delayed because they wanted to integrate some form of AI into it better, but the tech wasn't there initially. Not sure if that's full blown NPC interactions or just shit like traffic, though. |
|
Originally Posted By Towely: Not that it wouldn't make gaming awesome, but AI at that level is going to touch a lot more than gaming (obviously), and in a very bad way... No it won't. AI is not getting better. Any improvements on it are minute. The best it's doing is getting faster, and that's because they're throwing more compute at it. Eventually the monthly cost will be astronomical and put it out of the affordability range of nearly everyone. All LLMs are Markov Chains. They predict what the next word will be based upon context of the previous and the sum of their training data. The problem is that for most of the bots the training data is the internet, which is full of bad information. This is where hallucinations come from. The better you pare down the data, the better in accuracy it is, but now you start having a highly specialized program that becomes more limited in scope. After all, how are you going to handle errors? Say you have three enemies, but the LLM AI thinks it has 5 from a guess and issues invalid statements. Now your enemies are frozen because they've received an invalid statement in the game engine. Decision trees are one of the better options for easy AI, and how granular or branching you make the tree determines the depth of the AI. Goal oriented also works, something that FEAR did, see:  Building the AI of F.E.A.R. with Goal Oriented Action Planning | AI 101 And both of these do not require any external input. |
|
Originally Posted By Hayashi_Killian: No it won't. AI is not getting better. Any improvements on it are minute. The best it's doing is getting faster, and that's because they're throwing more compute at it. Eventually the monthly cost will be astronomical and put it out of the affordability range of nearly everyone. All LLMs are Markov Chains. They predict what the next word will be based upon context of the previous and the sum of their training data. The problem is that for most of the bots the training data is the internet, which is full of bad information. This is where hallucinations come from. The better you pare down the data, the better in accuracy it is, but now you start having a highly specialized program that becomes more limited in scope. After all, how are you going to handle errors? Say you have three enemies, but the LLM AI thinks it has 5 from a guess and issues invalid statements. Now your enemies are frozen because they've received an invalid statement in the game engine. Decision trees are one of the better options for easy AI, and how granular or branching you make the tree determines the depth of the AI. Goal oriented also works, something that FEAR did, see:  https://www.youtube.com/watch?v=PaOLBOuyswIAnd both of these do not require any external input. LLMs are absolutely not just Markov chains. They output the next token probablistitically, sure, but if they were simple Markov chains then there wouldn't be the ability to maintain any kind of context. There is a lot more going on than a simple Markov chain or AI would have been dead in the water out of the gate. To say that AI isn't getting better is verifiably false... I do think they are going to hit a hard wall, and I have my doubts that we will ever reach "true AI," vs. the extremely advanced machine learning we currently have, but to say what we currently have isn't improving daily is a wild claim. Hallucinations don't come from bad data from the internet. The solution to hallucinations isn't shrinking the scope, it's a design and architecture issue. More compute does more than just 'make it faster.' This has been tested and proven false... larger models show observable new behaviors/abilities and perform tasks that smaller models cannot, even if given infinite time. Cost is a real issue now, but it might not be the killshot we all hope it to be. Efficiency has been improving faster than demand has been increasing and the costs have been dropping every year. This same argument was made about the cloud. No one is saying that devs will simply plug their games into an LLM and let it ride, that isn't the point I was trying to make. Of course it will have to be locked into some kind of decision tree or GOAP to give any kind of direction. There are already examples of this out there and it's exactly how game devs will utilize AI. |
| considering Ai is completly killing gaming as a whole I would say the end of gaming is what Ai will do. |
callmenoshie: "saying that females have the potential to be "bat shit crazy" is like saying the sky has the potential to be blue."
XCRmonger: "I've seen German Shit Porn that was sexier."
XCRmonger: "I've seen German Shit Porn that was sexier."
|
Originally Posted By slappomatt: considering Ai is completly killing gaming as a whole I would say the end of gaming is what Ai will do. Don't get me wrong, if I could push a button and make AI disappear completely, I would. I'm just looking for a silver lining. When can show AI an old photograph and it can make a realistic video of the person in the photo, it stands to reason that AI is the easy button for making 3D charachter model for a game. As for dialog, if youve ever talked to an AI chatbot that has constraints like a "back story", they are still noticably "fake", but on a whole other level of interaction compared to typical NPCs. Imagine a play through of a game like Skyrim where instead of selecting from a set of responses, you could speak into a microphone and the NPC would respond naturally. I think that's possible already. It just needs to be integrated into games. But I am curious, not being argumentative. I'm just not in the loop. How is AI killing off video games? |
|
Originally Posted By Towely: LLMs are absolutely not just Markov chains. Originally Posted By Towely: LLMs are absolutely not just Markov chains. Originally Posted By Towely: They output the next token probablistitically You contradict yourself here. You're describing a Markov Chain. Originally Posted By Towely: but if they were simple Markov chains then there wouldn't be the ability to maintain any kind of context. They literally can't. Originally Posted By Towely: To say that AI isn't getting better is verifiably false... I do think they are going to hit a hard wall, and I have my doubts that we will ever reach "true AI," vs. the extremely advanced machine learning we currently have, but to say what we currently have isn't improving daily is a wild claim. A claim that's been made by a LOT of scientists for a while now, since as early as 2022. The only ones arguing the counter are techbros who are begging for please bro, just another 20bn so we can get AGI. Originally Posted By Towely: Hallucinations don't come from bad data from the internet. Now you're confidently talking out your ass. Hallucinations straight up happen because of the wide swathe of data used from Reddit and more and it draws conclusions based upon random data that it probabilisticly connects without understanding context (remember AI suggesting to add glue to your pasta sauce to thicken it, when it was meant to be a joke?). |
|
Originally Posted By Hayashi_Killian: You contradict yourself here. You're describing a Markov Chain. They literally can't. A claim that's been made by a LOT of scientists for a while now, since as early as 2022. The only ones arguing the counter are techbros who are begging for please bro, just another 20bn so we can get AGI. Now you're confidently talking out your ass. Hallucinations straight up happen because of the wide swathe of data used from Reddit and more and it draws conclusions based upon random data that it probabilisticly connects without understanding context (remember AI suggesting to add glue to your pasta sauce to thicken it, when it was meant to be a joke?). Originally Posted By Hayashi_Killian: Originally Posted By Towely: LLMs are absolutely not just Markov chains. Originally Posted By Towely: They output the next token probablistitically You contradict yourself here. You're describing a Markov Chain. Originally Posted By Towely: but if they were simple Markov chains then there wouldn't be the ability to maintain any kind of context. They literally can't. Originally Posted By Towely: To say that AI isn't getting better is verifiably false... I do think they are going to hit a hard wall, and I have my doubts that we will ever reach "true AI," vs. the extremely advanced machine learning we currently have, but to say what we currently have isn't improving daily is a wild claim. A claim that's been made by a LOT of scientists for a while now, since as early as 2022. The only ones arguing the counter are techbros who are begging for please bro, just another 20bn so we can get AGI. Originally Posted By Towely: Hallucinations don't come from bad data from the internet. Now you're confidently talking out your ass. Hallucinations straight up happen because of the wide swathe of data used from Reddit and more and it draws conclusions based upon random data that it probabilisticly connects without understanding context (remember AI suggesting to add glue to your pasta sauce to thicken it, when it was meant to be a joke?). That isn't contradicting anything. LLMs use probability... I never said that they only look at the last token, which is what markov chains do. LLMs absolutely look further back then that and do maintain context, and can 'understand' what previous information is relevant to the current prompt and what is not. You are simply confusing memory with context. They don't sustain memory like we do, sure, but to claim that my statement about context is wrong and that they "literally can't" is just flatly false.. Context is not memory, and it's absolutely using the former even though it doesn't have the latter. Hallucinations are 100% a training and model problem, not a bad data problem. Many AI models are trained to answer confidently no matter where their information falls outside the high confidence region. This is observable... This is why you can also 'fix' a hallucination by prompting the LLM to only provide answers it is confident in, and to instead say that it does not know if there isn't any data it can find that answers the question with high confidence. The reason hallucinations happen, overwhelmingly, is because LLMs are trained to always provide an answer confidently, so if it cannot find something that 'it is confident' is true, it will grab on to something that seems linguistically correct and state it as fact. Even a model trained on perfect data would hallucinate if it has been rewarded for fluency over admitting it "doesn't know." |
Back on the topic: AI Generated Games Just Got Eerily Real |
|
Crazy. So it sounds like within a few years nobody will be buying a video game, they will buy a subscription and tell the AI what kind of environment and story line they want to interact with, and everyone will just play in their own custom worlds. I imagine there will also be ways for multiple players to play online in the same environments. |
|
Originally Posted By DaGoose: There was a book series (Unincorporated Man series) that had the collapse of civilization around AI games that addicted most everyone to them to the point where they wouldn't eat. Plot twist: AI gaming cured the obesity epidemic 
|
"Makes me realize why there are warning labels on hemorrhoid cream about not using it orally" - Weomi
Save VA, Join or Donate to VCDL!
https://www.vcdl.org/Donate
Save VA, Join or Donate to VCDL!
https://www.vcdl.org/Donate
Joined:
May 2026
Posts:
2
EE: 0% (0)
FRA
|
Originally Posted By Jambalaya: How so? Don't get me wrong, if I could push a button and make AI disappear completely, I would. I'm just looking for a silver lining. When can show AI an old photograph and it can make a realistic video of the person in the photo, it stands to reason that AI is the easy button for making 3D charachter model for a game. As for dialog, if youve ever talked to an AI chatbot that has constraints like a "back story", they are still noticably "fake", but on a whole other level of interaction compared to typical NPCs. Imagine a play through of a game like Skyrim where instead of selecting from a set of responses, you could speak into a microphone and the NPC would respond naturally. I think that's possible already. It just needs to be integrated into games. But I am curious, not being argumentative. I'm just not in the loop. How is AI killing off video games? I think the fear is less “AI will instantly destroy gaming” and more that publishers will use it to cut corners everywhere possible. People already complain about unfinished games, repetitive open worlds, lazy writing, and copy paste content, so a lot of players worry AI could make that even worse if companies start mass generating dialogue, quests, textures, or voice acting just to save money. That said, I do agree there’s also huge potential if it’s used carefully. Stuff like more natural NPC conversations, smarter procedural events, or reactive worlds honestly sounds really cool to me. I think the problem is not the technology itself so much as whether studios use it to improve games or just pump out cheaper content faster. |
|
Originally Posted By Jambalaya: How so? Don't get me wrong, if I could push a button and make AI disappear completely, I would. I'm just looking for a silver lining. When can show AI an old photograph and it can make a realistic video of the person in the photo, it stands to reason that AI is the easy button for making 3D charachter model for a game. As for dialog, if youve ever talked to an AI chatbot that has constraints like a "back story", they are still noticably "fake", but on a whole other level of interaction compared to typical NPCs. Imagine a play through of a game like Skyrim where instead of selecting from a set of responses, you could speak into a microphone and the NPC would respond naturally. I think that's possible already. It just needs to be integrated into games. But I am curious, not being argumentative. I'm just not in the loop. How is AI killing off video games? edit. tried to insert a meme but this god forsaken shit new BBS we "upgraded" to still wont host a imageflip link. |
callmenoshie: "saying that females have the potential to be "bat shit crazy" is like saying the sky has the potential to be blue."
XCRmonger: "I've seen German Shit Porn that was sexier."
XCRmonger: "I've seen German Shit Porn that was sexier."
|
AI will lower the bar for game development. I've spent the last month working mostly in the evenings to build my own nuclear power plant simulator. I'm just a few weeks away from an initial release! The sim features high-fidelity physics of a simplified, single-loop pressurized water reactor plant. While it's a great tool for training brand-new nuclear engineers, the barrier to entry is incredibly low - it is designed for absolutely anyone to play and learn from. It includes a full training campaign that takes you from zero knowledge to a point where you can easily control the plant and understand its dynamics and the underlying physics. Best of all, it works entirely in the browser directly on my website. |
"Your enemy is never a villain in his own eyes. Keep this in mind; it may offer a way to make him your friend. If not, you can kill him without hate — and quickly." -Heinlein.
|
There’s been at least one AI (ChatGPT) follower out on Skyrim for three years now… https://www.nexusmods.com/skyrimspecialedition/mods/89931 And even programs that lets you add AI to any NPC in the game… https://www.nexusmods.com/skyrimspecialedition/mods/98631 https://www.nexusmods.com/skyrimspecialedition/mods/126330 Back many years ago they tried AI on vanilla NPCs in Skyrim for more realism during sandboxing, but the NPCs started doing strange things like going on quest and stealing from other NPCs including killing them for what they wanted.  Skyrim Ai Doesn't Just Talk Anymore...It ACTS! |
I have no useful skills, therefore I supervise.
Call sign:Radio Silence
Call sign:Radio Silence
|
Fire up chatgpt and ask it to DM a game of dungeons and dragons for you. It'll do a fairly good job out of the box. Give it better prompts and instructions, it'll do better. It'll still drift and hallucinate as the convo gets long. Use a better tool with local rules to save state and enforce that output matches pre-determined state and "lore", and it gets.... very interesting. Whoever makes a good DnD-style game engine to enforce continuity and has it connect to a decent AI backend for content, conversations, encounters, descriptions... is going to make bank. |
|
Originally Posted By RaptorFuel: AI will lower the bar for game development. I've spent the last month working mostly in the evenings to build my own nuclear power plant simulator. I'm just a few weeks away from an initial release! The sim features high-fidelity physics of a simplified, single-loop pressurized water reactor plant. While it's a great tool for training brand-new nuclear engineers, the barrier to entry is incredibly low - it is designed for absolutely anyone to play and learn from. It includes a full training campaign that takes you from zero knowledge to a point where you can easily control the plant and understand its dynamics and the underlying physics. Best of all, it works entirely in the browser directly on my website. Is it possible to create a meltdown and watch it burn all the way to the earth’s core while simultaneously radiating the entire planet thus creating CHUD’s that we have to fight with hand tools? Cause I’d seriously buy it if that were the case. |
|
Originally Posted By jerrwhy01: Is it possible to create a meltdown and watch it burn all the way to the earth's core while simultaneously radiating the entire planet thus creating CHUD's that we have to fight with hand tools? Cause I'd seriously buy it if that were the case. |
"Your enemy is never a villain in his own eyes. Keep this in mind; it may offer a way to make him your friend. If not, you can kill him without hate — and quickly." -Heinlein.
| AI has already fucked up gaming. Take Warzone for example. Ai monitors all chat communications in game and if you say any word that is offensive you get chat banned. You can say "fuck, Mother fucker, nigga" all day long but if you say "tranny, fag, or faggot" immediate chat ban. The stupid tranny and women developers have killed gaming. |
All politicians are evil, but some politicians are more evil than others.
Who is John Galt?
Who is John Galt?
|
Eventually.... There will not be much left for multiplayer games as it will require multiple players to agree to play the same game, but the AI will make it so easy to make branch-off variations to the end-user's specifications. Each variation will be full of more and more realistic AI players to compensate for the lack of mutual people players. It will be just like going to work. Me, a mindful thoughtful human, surrounded by a bunch of NPC's with clever little tricks to attempt to prove to me that they're people. I am a human, right? |
|
We're going to get a bunch of amazing games; the kind of stuff specific franchise fans could only dream of but would never actually be made prior to AI. However, the market will be so flooded with thousands of other new games that each of these great games will be impossible to find. The high cost of making great games will switch from production to advertising. AAA studios will still reign supreme, while only a handful of titles will be able to enter the spotlight due to getting lucky via independent streamers and content creators. I believe we're going to go back into the days of console-wars. |
Joined:
Dec 2022
Posts:
4952
EE: 0% (0)
MI, USA
|
Originally Posted By ArGyLe64: AI has already fucked up gaming. Take Warzone for example. Ai monitors all chat communications in game and if you say any word that is offensive you get chat banned. You can say "fuck, Mother fucker, nigga" all day long but if you say "tranny, fag, or faggot" immediate chat ban. The stupid tranny and women developers have killed gaming. What does that have to do with AI? |
|
Originally Posted By CTYC313: What does that have to do with AI? Originally Posted By CTYC313: Originally Posted By ArGyLe64: AI has already fucked up gaming. Take Warzone for example. Ai monitors all chat communications in game and if you say any word that is offensive you get chat banned. You can say "fuck, Mother fucker, nigga" all day long but if you say "tranny, fag, or faggot" immediate chat ban. The stupid tranny and women developers have killed gaming. What does that have to do with AI? From personal experience, I'm deep into the Battlefield 6 and Marathon communities. The amount of people getting banned/suspended in those games for ridiculous reasons is insane. It's mostly automated moderation carried out by AI. |
Joined:
May 2026
Posts:
5
EE: 0% (0)
FRA
|
Originally Posted By Towely: You are simply confusing memory with context. They don't sustain memory like we do, sure, but to claim that my statement about context is wrong and that they "literally can't" is just flatly false.. Context is not memory, and it's absolutely using the former even though it doesn't have the latter. Hallucinations are 100% a training and model problem, not a bad data problem. Many AI models are trained to answer confidently no matter where their information falls outside the high confidence region. This is observable... This is why you can also 'fix' a hallucination by prompting the LLM to only provide answers it is confident in, and to instead say that it does not know if there isn't any data it can find that answers the question with high confidence. The reason hallucinations happen, overwhelmingly, is because LLMs are trained to always provide an answer confidently, so if it cannot find something that 'it is confident' is true, it will grab on to something that seems linguistically correct and state it as fact. Even a model trained on perfect data would hallucinate if it has been rewarded for fluency over admitting it "doesn't know." I think the distinction between context and memory is an important one, and I agree that people often mix the two up when talking about LLMs. The hallucination point makes sense too. Even with good training data, if the model is pushed to give an answer instead of admitting it doesn’t know, it can still end up producing something that sounds convincing but is completely wrong. That’s probably one of the biggest things that needs to improve before people can really trust AI for more serious applications. |
|
Originally Posted By leonmarkos: I think the distinction between context and memory is an important one, and I agree that people often mix the two up when talking about LLMs. The hallucination point makes sense too. Even with good training data, if the model is pushed to give an answer instead of admitting it doesn’t know, it can still end up producing something that sounds convincing but is completely wrong. That’s probably one of the biggest things that needs to improve before people can really trust AI for more serious applications. Originally Posted By leonmarkos: Originally Posted By Towely: You are simply confusing memory with context. They don't sustain memory like we do, sure, but to claim that my statement about context is wrong and that they "literally can't" is just flatly false.. Context is not memory, and it's absolutely using the former even though it doesn't have the latter. Hallucinations are 100% a training and model problem, not a bad data problem. Many AI models are trained to answer confidently no matter where their information falls outside the high confidence region. This is observable... This is why you can also 'fix' a hallucination by prompting the LLM to only provide answers it is confident in, and to instead say that it does not know if there isn't any data it can find that answers the question with high confidence. The reason hallucinations happen, overwhelmingly, is because LLMs are trained to always provide an answer confidently, so if it cannot find something that 'it is confident' is true, it will grab on to something that seems linguistically correct and state it as fact. Even a model trained on perfect data would hallucinate if it has been rewarded for fluency over admitting it "doesn't know." I think the distinction between context and memory is an important one, and I agree that people often mix the two up when talking about LLMs. The hallucination point makes sense too. Even with good training data, if the model is pushed to give an answer instead of admitting it doesn’t know, it can still end up producing something that sounds convincing but is completely wrong. That’s probably one of the biggest things that needs to improve before people can really trust AI for more serious applications. Yeah, LLMs by themselves have a lot of failure points. A good AI gaming system will have a "regular old school" program to manage permanent state tracking, sanity, rules, etc., and appropriately choose what to feed to each call to the LLM to generate a relevant context and generate the context. When you just have an LLM run a game directly, it goes off the rails fast and hard. |