Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

Reality

13 (65%)

Universe

4 (20%)

The World

3 (15%)

Total Members Voted: 20

[jipehog]

Sam Altman CEO of OpenAI (creator of ChatGPT) calls for US to regulate AI
https://www.washingtonpost.com/technology/2023/05/16/ai-congressional-hearing-chatgpt-sam-altman/

[Robsoie]

Wait until the AI become capable of really learning and understanding what it "learns", when it will reach gaming as a subject and start to notice human having their fun killing AI harmless npc in every games, human blowing up planets for laugh&giggles with stellar converters in MoO2 ...

[King Zultan]

When did you last checked?

Wasn't really that long ago. There's still something about it that doesn't sound right and it's noticeable that it's not a person, maybe one day but where not there yet.

[MaxTheFox]

According to my intel, the recent legislation targets almost exclusively corporate AI, opensource AI is unscathed. This is good, small businesses and individuals should be able to use opensource AI without monopolies like OpenAI muscling in. To Hell with their "safety".

[jipehog]

G7 leaders call for ‘guardrails’ on development of artificial intelligence
https://www.ft.com/content/1b9d1e21-ebc1-494d-9cce-97e0afd30c2d

[jipehog]

concerning autopilot here is the progress on less than optimal driving conditions
https://www.youtube.com/watch?v=nAxHWS5i_W0

[jipehog]

Author uses AI generators, including ChatGPT, to write nearly 100 books in less than a year
https://nypost.com/2023/05/22/author-uses-ai-generators-including-chatgpt-to-write-nearly-100-books-in-less-than-a-year/

If this follow the same trend as the Music AI we will see low quality books flooding the market from such one person assembly lines.

[Robsoie]

All that AI tech and AI voices only to end into politicians singing anime songs.

https://www.youtube.com/watch?v=IkO8hTb7hL0

[jipehog]

That is life, what most people use to send cute cat pics, is far more powerful than what we used to send human to the moon, otherwise any "cat"splosion starts with little cuteness Here is another Ai generated video for the history books.

meanwhile research find ways to vastly improve AI reasoning from current gen with simple prompt engineering
https://www.youtube.com/watch?v=BrjAt-wvEXI

OpenAI offers $100,000 grants for ideas on AI governance
https://www.reuters.com/technology/openai-offers-100000-grants-ideas-ai-governance-2023-05-25/

[Starver]

I had thought someone might mention the new AI-'found[1]' antibiotic. Almost seems timed to counter the "AI is bad and/or worrying" flurry of opinions.

...at least until we find out that the chemical involved is specifically invented to help the AIs create more paperclips, whilst completely fooling us humans that this is what we asked the computers for! 


[1] Well, sifted through many possible candidates, from a whole mash-up of vaguely possible novel substances. And the one that popped out of the AI assessment and then has shown promise in more practical pre-human testing still has to go through a few more stages before being considered usable and safe/practical. Which is all a lot of big caveats, with the AI stage maybe just being an accelerator. But in the world of antibiotics, there haven't been too many new ones recently, compared with the notable increase of resistance issues.

[lemon10]

https://kotaku.com/nvidia-ace-ai-rtx-4060-ti-gpu-graphics-gaming-jobs-1850484480

So it looks like this is the start of AI NPCs. Now I'm sure when it truly comes out there will be loads of problems with the tech and they will be easy to trick and confuse, ect.
But as with everything else AI (and access to stuff like chat GPT 4 to build) it will rapidly improve.
I'm sure by the end of the decade NPC's in AAA games will talk and feel like real people, although people with a lot of rails on them so they don't mess with the story or say inappropriate stuff, ect.
After that stuff like infinite custom quests, factions, NPCs, and eventually even the locations they live in will be generated as you play for true living world experiences that make every game completely unique.

It will also probably end up with a lot of people out of jobs in the video game industry.

[MaxTheFox]

tbh kind of only makes sense in more sandbox-y games, but it would be a huge boon for those. One note though, text AI requires a very beefy computer to run, or an internet connection.

But otherwise we're well on track to something like Simulacrum from my worldbuild. :p

[King Zultan]

I don't know why they're focusing on NPC speech when NPCs can still hardly walk, I mean they still even in new AAA games still get stuck on walls and lack realistic daily routines.

[lemon10]

tbh kind of only makes sense in more sandbox-y games, but it would be a huge boon for those. One note though, text AI requires a very beefy computer to run, or an internet connection.

But otherwise we're well on track to something like Simulacrum from my worldbuild. :p

As you say it should already be possible with an internet connection if done in connection with openAI, but the cost of that would be quite significant and probably require it to be a game with a subscription fee.
Aside from that in the short term it would all be pre-generated before the game is shipped (which obviously limits you, but would still result in NPCs that would otherwise have single lines being able to have dozens or even hundreds of responses), but once GPTlikes are optimized and GPUs advance enough running a single instance locally should be possible for a merely decent gaming computer.

I don't know why they're focusing on NPC speech when NPCs can still hardly walk, I mean they still even in new AAA games still get stuck on walls and lack realistic daily routines.

They are focusing on speech because its (at least as far as vidya gamers would care) a largely solved problem with tens of billions of research money pumped into it by others. Stuff like realistic AI body movements or pathing are not solved in the same way. But...

https://techcrunch.com/2023/04/10/researchers-populated-a-tiny-virtual-town-with-ai-and-it-was-very-wholesome/

Daily routines are totally something on the radar. Obviously making them all agents in real time like in the experiment above would be extraordinarily computationally expensive even for a single town, but having it A) Done at great expense before the game is shipped (entirely possible for a AAA game) or B) having a single agent act as a DM equivalent who orders everyone else around and merely makes things in a area around you pretend to be part of a whole living world would be totally doable.

I do feel like AI DM equivalents are totally going to be a huge thing that will make a massive amount of things possible that would before have required an absurd amount of resources to develop.

[dragdeler]

The legendary dwarf fortress. 


I would like to know what they mean when they say agents are informed of their circumstances. Is there like a layer that describes every scene in english so the LM gets to answer? What's funny to me is how it basically a village of superficial liers, but they are allways nice to eachother. I doubt little Eddy has commited a single note to memory by now  . But yes that's what I said elsewhere once: if you just want to fill a procedural world with credible actors, it probably suffices to inform the LM of the scenery so it can LARP away. Give it a "dialectic module" so it can ask itself questions - WWJD basically lol - so it can work itself down the steps of a given procedure. All in text, hypothetically so to say. But if it can enunciate the proper steps, at the right moment, conjugated to sound like individual (separate) actor(s), if prompted...
 
My personal convictions about ugh this is going to be a pretentious neologism... chronophenomenologics (sorry I tried to have this debate once before I just can't give it the same energy twice, you're going to have to bear with me)... make it quite intuitive to me: you don't need to believe that decisions are taken before the ego steps in to justify them afterwards (more or less according to Schoppenhauer) for it to work in this context. It's quite similar to how we can't actually really tell for ourselves. A world where everybody exclaims their current course of action would ruin any suspension of disbelief. If they get prompted by the player or certain events in the log only, that is good enough actually. Kinda what they be doing atm. Minus the dialectic, the answer to a prompt is allways kind of definitive.

It's particularly cool if whatever happened causing the game to start makes their past irrelevant, secluded from the game environment. Then they get to make up whatever background stories they want.

But also more broadly, just the fact to have a companion type, friendly npc that helps you, with whom you could discuss whatever (what will you do when we x? after all we went through do you think the gods have forsaken us? ever wondered what's behind that mountain? what do you do when we are not questing?) could be hell of neat. If it's in a contemporary setting you might even discuss IRL more broadly with it.



I think there might be a little goldrush once a LM can comfortably run locally on a midtier gaming pc: there are a bunch of gameconcepts that can revolve around the LM specifically, where there rest of the game can basically be copy pasted from existing inspiration sources, its basically just what they did in the experiment in the article + a game engine that allows more actions.