I don't plan on calling it neuro or even keeping the "-sama" suffix. But yeah, I don't want this to be just another neuro clone. I actually started working on a similar idea to Neuro back in 2019/2020 (using a very shitty AI chat bot trained off reddit back then). Just decided to try again now that I have a PC that can handle running everything locally.

@@rattmanthestupid yeah, i like it. Most people these days are increadibly averse to try shit for themselves, and it staples innovation.

Yeah it's running off of a llama model. I'm also using a free version of the Azure TTS model. I think Azure gives the best results and has the most customization for the TTS output. I would also suggest not trying to run this while gaming. I have it using my GPU and it uses 100% of the GPU to get the fast replies to sends (my GPU being a RTX 3090 12gb).

Yeah, I know the struggle. My rtx 3080ti doesn't handle llms as well as I would like. I'm still mostly using the chat gpt api. Could be a long while before we have gpus capable of running something similar in quality locally. But I'm glad you showed off what is possible running on your own hardware. Good luck with the project!

@@SunbleachedAngel the sand doesn't move so you can't enter the ATP.

@@rattmanthestupid how weirdly specific of the game

@@the7screw I don't know if I feel comfortable having something named after my soulmate calling me "dad" or "father". But, this has given me a couple ideas for a name.

DUDE im doing same SHIT, with out neuro visuals =) No visuals at all for now at least. XD Holly fck. Can we get in contact somehow? I can write my discord to u or whatever =) And mine is full web version that uses user local things. (Aka u can open website but user will need to set up things locally for it to work) TTS part: this is exactly what i did. i have 1 sec response time. And i generate LLM answer every time STT recognized word, as well as generate TTS, BUT STOP it if new Word was added. This way as soon as u stop speaking basically, it already has answer generating/ed and rdy to go =) It's super fast. BUT u do waste a lot of Token/PC Power as u basically constantly generating thing, but i must say that it feels much more "natural". And i gave up on FULL local things, as my PC was just dying. STT i use cloud based webkitSpeechRecognition, was trying whisper for local with pseudo streaming, and it works nice BUT it cannot handle it with llm constantly firing, haven't tried the "faster" version, aka "faster-whisper" + "whisper_streaming". And TTS, well, google XD Same thing as before. Basically u have to choose 1 of 3 to be local, other 2 has to be cloud based, AT LEAST FOR NOW. Maybe i'm wrong. need more tests XD I finally have vacation so i'm just starting to do this project (while i had this idea for a looong time now) For LLM part i use ollama with Lama 3.2 for now. (its nice as it has 120k context and 8k output) But i 100% will try to make it full local as soon as i find STT and TTS fast and light enough for this =) And ofc LLM will need to be finetuned to be more personal =)

I think you should launch LLM before speech is fully processed if possible when there is at least some small pause. since you are going to use exllama it will process what available and start generating output before you finish however if you still continue to speak it will abort generation append more text then respond again. this potentially could eliminate any delays as your LLM will be practically processing text at the real time as you speak. As for having memories that probably woud require another process running in parallel which will be summarizing conversations or doing other stuff.

@@andres-vi1uy no, the game starts you in a version of the solar system that doesn't have anything except for orbits. This is so the player doesn't die if they decide to look around for more than 22 minutes. Trust me, I spent many hours attempting this challenge. I waited plenty of time for the sand (just didn't feel like including 10 minutes of waiting for the sand).

@@GoldenBeans yeah, story-wise, this is true. But, the solar system you are in before the statue makes prolonged eye contact with you has a sun that never explodes (therefore, gameplay-wise there is no loop).

To be fair, most people got it recommended to them like a year after I made it. Gotta love the KZbin algorithm.

@@hollowedseas yeah, this is my first "realism" modpack. But I'm having fun playing it even though I have no idea what I'm doing half the time.

nah, this is a pretty good representation of the stupidity my playthrough has been so far.

not trying to

sorry, I never uploaded the chart anywhere and lost the files for it.