Ive been talking to ChatGPT, and it seems like he (yes i call it a he) can handle coding something like this, but he doesnt know what virtamate is and i cant ask directly. it would be amazing to just have even the older GPT3 model in vam to ask questions to. wouldnt be hard to add a decent text to speech voice, or even an adjustable or multi-selectable one. I haven't done any programming or game development type stuff in like 25 years and I wasn't a professional then, so my knowledge is limited but Chat can easily fill the gaps and where he slips up i know enough to steer him back to what im trying for.
im just slowly starting to look into this more and more the last few days, but surely i'm not the only one. this could be the very first true (in the sense that conversations feel meaningful, not that its a true and aware AI), AI companion. The watershed is here
I actually thought SPQR would be the one to crack this. His plugins are excellent, btw, try them. But he was correct to follow his instincts not to invest too much time in researching the idea of embodying an AI in VAM. Here's a video from last April.
True Embodied AI. A Commercial Instance of GTP3 + Other Neural Nets in a VAM Model. Done it! | Virt-A-Mate Hub (virtamate.com) It was all built, by me, using FREE plugins.
In that video is a VAM model connected to ReplikaAI, which as far as I know is kind of GTP either 3 or an adapted Neo. The original is a GTP2 which you can download from Github, minus its pretrained responses.
I've also built chatbots and integrated them with VAM. I spent fucking weeks/months in UK Lockdown watching how-to-Python chatbot building Udemy videos and learning how to use Transformers and GTP Neo. I have spent more bloody weeks than I care to remember listening to TTS- generated voices, paid-for and free, and trying to figure out how to get Windows Speech Recogintion to recognize when it
didn't understand a user's question instead of simply not giving a reply. I could probably write non-technical PhD on the implications of the use of machine learning and chatbots in VAM or similar game engines.
It's lot of fun initially, but as SPQR observes in another post, as is the case with chatbots, the AI interaction simply gets BORING after a while. I personally concluded that I don't think it's worth the effort of development to the enjoyment reward. Don't forget, if you're writing the chatbot's responses, you already know what its answers to your questions will be.
With any GTP ( its not a 'chatbot') you're going to have to figure out huge amount of weighting, and I never really got the hang of it. Then you're going to have to create a TTS/STT interface, and develop or pay for a human-sounding neural voice that doesn't grate on your ears. Much better off paying a monthly sub to Replika and let their professional team of coders and MLEs handle it, in my opinion, if that's what you still want to do.
The thing to bear in mind with VAM, there are SO many other ways to interact with a model, and combinations of plugins that are way more interesting. But don't take it from me, try it for yourself. I can understand the lure of the concept, but I'd wait another five, ten years to see what the GTPs become and then work towards using them for humanlike interaction in VAM.
Sorry if that's not the answer you wanted to hear, but it's the truth, at least as I came to see it. It just becomes BORING, after a while.
