ChatGPT powered Yandere AI Girlfriend Simulator

[atarumoroboshi18]

Free game, but requires an OpenAI API Key

This is a crazy one. Essentially, you are stuck in a room with a crazy Yandere who calls herself your girlfriend and your goal is to escape the room. The twist is, she's powered by ChatGPT meaning no two playthroughs will ever be the same, chat about anything you want. Have conversations, compliment her, make small talk, be angry, be scared, be meta, etc. and using ChatGPT, you'll be making your own story that can be completely different every single time.

https://helixngc7293.itch.io/yandere-ai-girlfriend-simulator

 

[RanceJustice]

This is interesting, though it is an example of the increasing concern for how AI will be integrated into future projects. Both ChatGPT and the Azure voice services are cloud based solutions that require API keys integrated in order to work - its subscription based. Now, this is just an indie title and a more professional game would likely spend a considerable amount to bind a particular API access key for either ChatGPT or the Azure voice to be "built in" without requiring user configuration, its still dependent on a handful of proprietary AI platforms accessed remotely and paid as a service. This needs to be stopped and the more it is normalized the worse things will become.

 

[LukeTbk]

It saved potentially so much money when making the game that it does not need to be seen has something to be stopped.

Like anything that is cloud base or would be hard to run in a virtual machine, worry about longevity and so on obviously apply.

Big pro game (if they do not need the carachter in the game to up to date with the news, like say Red Dead Redemption 3), could train their model, build it and ship it with the game to run locally on your computer I think, specially with how good cpu-gpu are getting at those workload, a bit like:
https://github.com/tatsu-lab/stanford_alpaca/tree/main
https://github.com/oobabooga/text-generation-webui

https://www.tomshardware.com/news/running-your-own-chatbot-on-a-single-gpu
Getting the models isn't too difficult at least, but they can be very large. LLaMa-13b for example consists of 36.3 GiB download for the main data(opens in new tab), and then another 6.5 GiB for the pre-quantized 4-bit model(opens in new tab). Do you have a graphics card with 24GB of VRAM and 64GB of system memory? Then the 30 billion parameter model(opens in new tab) is only a 75.7 GiB download, and another 15.7 GiB for the 4-bit stuff. There's even a 65 billion parameter model

A game would not need its characther to know everything around the world and be chatgpt-4 good in every domain, I could be all wrong but giant budget game could make their own training model that run locally, but obviously would be potentially tempted to run it on their server, but I am not sure if it something to be stopped (why ?), imagine what it open in term of NPC capability, cities and so on.

Why would it matter if a Rockstar game use something proprietary, their game engine is, their audio engine is, it is a paid-closed product.

I think it could be surprising (it was to me) has we are used to very large data set (or think that the very large data set used for their training will create a somewhat really large dataset for the final product), but apparently the largest GPT-3 model (175 billion parameter) was only 700GB.

Fit on some Iphone and considering how small of a fee it is to use openAI (they could be operating on a giant loss obviously, but still) a single simpe request like NPC dialogue could probably run on a regular cpu-gpu on a model that fit on a regular game install (I could be saying absolute non sense)

 

[RanceJustice]

It saved potentially so much money when making the game that it does not need to be seen has something to be stopped.

Like anything that is cloud base or would be hard to run in a virtual machine, worry about longevity and so on obviously apply.

Big pro game (if they do not need the carachter in the game to up to date with the news, like say Red Dead Redemption 3), could train their model, build it and ship it with the game to run locally on your computer I think, specially with how good cpu-gpu are getting at those workload, a bit like:
https://github.com/tatsu-lab/stanford_alpaca/tree/main
https://github.com/oobabooga/text-generation-webui

https://www.tomshardware.com/news/running-your-own-chatbot-on-a-single-gpu
Getting the models isn't too difficult at least, but they can be very large. LLaMa-13b for example consists of 36.3 GiB download for the main data(opens in new tab), and then another 6.5 GiB for the pre-quantized 4-bit model(opens in new tab). Do you have a graphics card with 24GB of VRAM and 64GB of system memory? Then the 30 billion parameter model(opens in new tab) is only a 75.7 GiB download, and another 15.7 GiB for the 4-bit stuff. There's even a 65 billion parameter model
A game would not need its characther to know everything around the world and be chatgpt-4 good , I could be all wrong but giant budget game could make their own training model that run locally, but obviously would be potentially tempted to run it on their server, but I am not sure if it something to be stopped (why ?), imagine what it open in term of NPC capability, cities and so on.

Why would it matter if a Rockstar game use something proprietary, their game engine is, their audio engine is, it is a paid-closed product.

Putting aside that "it saves us/makes money, of course we should do it!" is one of the reason that some of the most exploitative elements in gaming (among elsewhere) are getting worse and worse such as gacha monetization and the like..

If they managed to have every built entirely on your PC that would be an improvement, but the issue is the accessibility of those models and their quality vs the "big name" cloud based/API connected ones. Now maybe Alpaca will finally start to move things in the right direction, but the concern is that outside of Stable Diffusion, the "class leading" AI models and even training data are mostly kept to themselves and only accessed by 3rd parties like t his one. Elevenlabs AI voice cloning, CharacterAI, ChatGPT, and many others work this way. This also has the benefit of all the elements interacting with them - including this very game - become options in training the model, which further creates a rift between what "big names" can do and what hobbyist options can provide. So sure, there will be some hobbyist version, but much the same way that you can have a social network without the userbase of and data from Facebook, Facebook has much more advanced data, metadata, and algorithms to make determinations thereby. Every one of those kids asking ChatGPT to do their homework or answer their texts becomes training data which your homebrewed alternative will not be able to source. Or to put it more succinctly, if you're making a game and you want AI characters, scripts or the like why go through tens or hundreds of gigabytes or more of training data and bend it to youirself to get an AI character response that is baseline, when ChatGPT will give you what is advertised to be the very best AI subsystem for your game at a license cost of only $X per copy and a helpful API and SDK meant to integrate it into their project?

A handful of those who care a lot about player autonomy, are running on a shoestring budget and are maker focused etc...may end up taking the open DIY path, but especially if it appears the gulf between the big name "AI as a service" provider fidelity/quality and the DIY versions continues to grow, many will not - all they care about is the smoothest way to get that functonality into their game. I absolutely advocate for more open AI solutions and training data availability, but given the huge amounts of money, first-mover or network effects, and lack of constraints in a fast moving industry I don't anticipate this to end up well when there's yet another oligopoly of AI companies to whom near everyone must kowtow. For the past 30 years or so we've been told that all these amazing technical developments are going to really make things better and accessible and universal etc.. and in the vast majority of cases they've been increasing intrusions on privacy, data sovereignty, autonomy, security, value, and control when they're in the hands of a handful of overlords ensuring their proprietary services are the only real options; its great for their bottom line, and everyone else just has to deal with it.. I don't see any change in this regard with AI unless there is some sort of hard limitation to put an end to this before one or a few projects are able to get out far enough ahead of everything else exploiting the lack of regulation.; There are other dimensions of the discussion , but unless things change soon I don't have much faith that the benefits of AI models will openly available as opposed to being controlled and doled out by a handful of digital oligarchs.

 

[socK]

There are open source LLMs that are pretty good, but they're still heavy to run locally. They use enough memory that you'd probably need 12+gb of VRAM at absolute minimum.

The CPU version probably wouldn't work reliably unless you have 64gb.

 

[LukeTbk]

I feel it is normal to hate game that connect with no benefit (like for anti-piracy version), but if it is for value (like world of warcarft or CS:Go to play with other humans) it is not much of an issue. Connecting to get to play with other humans was not that far from paying-connecting to super NPC.

If it is to have NPC that goes world beyond what they could do otherwise (or to make game that would not have been realistic to make for a very small team)

 

[staknhalo]

ChatGPT powered Yandere AI Girlfriend Simulator​

1) you people disgust me

2)

 

[Gulkor]

Welcome to Pluto in Aquarius. The next 4 months will show us what next 20 years will be like.