Home
Artists
Search Recent Random
Posts
Search DMs Popular Hash Lookup Tags Random
Importer
Import FAQ
Register
Partychan Telegram ThePornDude
Home Artists Posts Import Register

Why is there a $15 tier? Why does AI Chat cost so much? (Patreon)

Published:
2023-07-11 14:50:26
Imported:
2023-07

Content

It currently costs us around (rough estimate) $1 per hour per user in server fees (with a flat-rate minimum cost of about $2,500 per month in our current configuration). Hopefully we can decrease these costs over time.

With the added $7 per month that we are charging for AI, we will break even if there are enough users, plus if people use the service 7 hours or less per month on average. This is not taking into account development expenses, only server fees.

In short, at least for a while, we will be losing money, not making money, by introducing new AI features into vamX.

https://www.analyticsvidhya.com/blog/2023/05/how-openai-lost-half-a-billion-dollars-due-to-chatgpt/

Estimates are that it costs OpenAI (the makers of ChatGPT) $0.36 per question you ask: https://www.govtech.com/question-of-the-day/how-much-does-it-cost-to-run-chatgpt-per-day

It takes a lot of server power to run "AI" models (LLMs) quickly. The server power must come from expensive GPU graphics cards on hosted computers. We host 5 different LLMs separated into 3 servers in order to have AI Chat work quickly. We are also working on creating a backup set of another 3 servers to further distribute the load and provide both US (current) and Europe (coming soon) servers.

As the technology progresses, if server costs lower dramatically, we will offer lower pricing tiers for the service.

Feel free to stick with the standard Patron or VIP tiers, once we lift our heads up from this AI craziness, and even somewhat during it, we'll be adding new non-AI features!

If you want to take a journey and join us in exploring what cool stuff can be done with AI, check out the help docs for the new vamX AI Chat features and join the new VIP + AI + ChatGPT tier.

Comments

Rolo

excellent work friend VAMx, I really admire all your effort in giving this great advance to the VR platform with AI. And of course the new voice commands that will be added in the new version 1.31 are excellent. Thank you so much.

james ceurvels

Feel free to stick with the standard Patron or VIP tiers, where new non-AI features will be added, ------>as soon as we have the time.<---- so what does that mean once evry 2 months for us cheap guys. i get you guys are highly exited aboit ai and chat with the ai but don't leave us haning for the ones who don'thave that kind of money.

w

Good work on the new AI chat

vamx
>>112678025

Basically, development can only go to one place at a time, so development towards AI applications is instead of development in other directions. On the other hand, AI applications won't take up a huge amount of time forever, rather, like other features, once this gets settled it'll take up very little time, so right now it will delay other releases, as this is the main new feature, but later it will just be a feature that is mostly done, and with only occasional improvements, etc.

Anonymous

Suggestion to ease up on the cost is to make a script that we can edit that allows us to use our own openai account and also somehow integrate elevenlabs into it also. Would be cool if you guys are able to allow different uses of LLM like pygmalion... if that's even possible... would really make this game interesting.

Anonymous

think there is also open source that also allow you to talk to your ai via voice. .......

Anonymous

ooh and if your familiar with ooba or silly tavern ... allow us to create our own ai personality !!

Anonymous

Love the work / project! It is fun seeing this evolve. I'm sure this was discussed... What about individuals using their own OpenAI API key, but then again...if you're thinking revenue stream??? I'm rambling...nonetheless THANK YOU!

Anonymous

The AI stuff is super exciting! It's not quite in my budget atm, but the cost is totally understandable. I'll be keeping an eye on this. Maybe once it's had some time to mature a bit then I'll be in a more comfortable budget zone. Will 1.31 be released to lower tiers later? Without the AI chat I mean.

vamx
>>112715226

The issue with using your own OpenAI API key are: 1) She will often stop the chat when it gets too sexual, probably just the moment you really don't want the chat to stop. 2) You would only get a text response, it also requires servers so that she speaks the reply, and also so that the voice data is processed and turned into text that OpenAI (ChatGPT) understands. We are also running language models to understand the actions. So it's quite easy to have ChatGPT running, but you could do that anyway in a browser in VaM. But if you want her to be talking to you, and you want to talk to her, and you want some action/response, you still need part of this system running. It might be possible if you have a 4090 card to run Virt-a-Mate and some of the rest of this on your local machine, but if so, you still run into issue #1. Thoughts?

vamx
>>112715916

Yes. I wanted to get 1.31 out really soon, for the ChatAI features, but it also contains a HUGE set of additional voice commands which can work without ChatAI. So it will probably release in a week or two to the VIP tier, as usual.

vamx
>>112713162

Yes, you can actually create your own personality already, just by pasting it in to the top Context box. You can edit it to make any personality in the format shown in that box, or paste in .json files (with curly bracket type format { ... } ) created for silly tavern.

vamx
>>112713064

Yes, but it still costs money / resources to run this, but not as much. See my 1) 2) response above.

vamx
>>112712993

I think the responses are better than pygmalion, but I'll be interested to hear feedback once people are trying it out. See my 1) 2) above, but basically OpenAI doesn't really support good NSFW chat, so it will ditch you right when you need it to continue. Also elevenlabs is pretty expensive, you'll quickly end up paying the $7 per month this way, just with OpenAI and elevenlabs, and get a much worse service, I think? This would be without an AI for analyzing actions, responses, and without speech recognition (although you could probably get Microsoft speech recognition).

Anonymous
>>112715226

Wow didn't expect the response and insight, thank you. Obviously there were many other elements I didn't consider. Individual overhead as you laid out most likely isn't feasible en mass. Plus sounds like you have the server stack in play too. Looking forward to further development!

kaiho leung

Can I ask how much content of future updates will relate to AI? For example, will 1.31 be all about AI updates and afterward? Will there be hard to separate into two versions for tiers in the future(AI free version/AI version)?

vamx
>>112804013

Like with most vamX features, all efforts are focused primarily on the main new feature, but then, once it's done, efforts move along to other features. So once the main important AI features have been finished, we will return to our (huge) non-AI pending features list. Feel free to suggest improvements. Right now, 95% of our effort is in seeing what cool stuff we can do with AI, and this is likely to continue for about a month. It's not at all hard to separate versions. There will just be an AI button, but if you press it, you won't be able to use it much unless you were to enter the password. In fact, we will release vamX 1.31 for VIP users soon (with non-AI upgrades, mostly around voice commands).