So, generative AI really took off recently, right? Almost every company is jumping on the AI model or AI-driven software bandwagon.
I've noticed that data and its validation are major roadblocks in this space, especially for smaller companies. I keep thinking that blockchain could be the solution. My idea goes beyond just another "AI coin" (which is basically just a fancy term for crap coins).
What I'm thinking is to create a platform where people can generate content with AI (starting with images, probably) and earn scores based on their creations. These scores would eventually convert into tokens (similar to what Hamster or NOT coin are doing). There's also a component where users validate good prompts with images or good Q&A pairs.
The end goal here is:
1. Build a solid dataset for further training (think open-source datasets and models)
2. Allow people to earn from messing around with AI
3. Create a potentially valuable altcoin.
I've considered some possible downsides and have ideas on that, but first, I'm eager to hear what everyone here thinks.
While that sounds good, you have to invest on this project alone of yours. Without monetary support from yourself, you won't get any support from other investors that sees the goal of what you're doing here.
It's a good idea that you'll make people generate any related AI things like images but, that's where it will be abused by people who sees that there's a real money in it.
This is not a sustainable idea and even you create your own altcoin about it, it's bound to fall and lose its value.
I have some anti-abuse mechanisms in mind, like making people sign up through telegram or something like that which is not easy to make "millions of accounts" (Today I saw on X that with some mail servers, you can get ChatGPT Plus for free for a long time, and I am talking like that type of abuse) and also, there must be something like this, generations give the user least amount of money and validations and suggestions give them more. Also rate limits work to some extent.
About investments, I have a friend who actually showed interest.
Sign up is a normal thing but asking them to do it on different platforms will just make them tired and uninterested. But whichever you think is a good idea, free to try it.
That's nice if there's one that's interested to invest.
It all starts with people whom trusting you with your idea but as I've said, many have I seen that they failed.
Oh maybe you mean a gamification system, I call it PoC (Proof of Contribution), this is familiar and I have found various AIs by implementing a point system that can later become a requirement to get tokens from the project. Then the AI-based tokens will be used as gas every time you generate an image or video. I once managed to get a reward and profit of $450 in that type of AI project.
Example of a project I'm working on
The tricky part is how this token is going to get value on the market. You are going to distribute your token to users, so there is going to be a lot of supply available. Many people willing to sell their tokens to generate profit in fiat currency or other solid cryptocurrencies. And who is going to buy it? Where does demand is going to come from?
AI platforms don't pay people to become users. It's quite the opposite: users have to pay in order to make use of the AI services, like Midjourney and Sora. And it's not cheap. So, how could the inverse logic work in this situation?
The idea of Generate-to-Earn isn't new, some projects that already do it are NFPrompt (NFP) & Bittensor (TAO). I don't really understand the full concept, but ultimately it's just about making money from prompts. I personally think this idea is great, because the project could generate big data from the AI training conducted by users. The company benefits & so do the users.
Well, it can be also like a mini app or part of another platform. Where people already made accounts. And about the friend, yes. People who trust you are important and this is why I'm trying to validate the idea first.
Yes, exactly. Gamifying the idea of using Gen AI, data labeling or something similar. Exactly the point. However, I have a good infrastructure which is just turned off and I want to power it up again.
Well small and medium AI companies who care about development of the systems on their own (neural nets, models, finetunes, etc) usually struggle with different problems. The biggest one for my previous startup was A good image dataset which has no copyright problems so we ended up paying artists, photographers, graphic desingers, etc. to create food for our Stable Diffusion/FLUX fine tunes. But later we found another problem which was "labeling the very data we've got in form of prompts". See? People are providing their creativity and get paid for that. Imagine it like World Coin but not evil .
But about the distribution, it is still a big problem. My friend whom I mentioned, says it's better to give people TON or other affordable low-fee well-established coins and later, start to make a token of our own. I have to take it to my consideration.
Exactly. AI projects have a data problem and it can solve them. I may be writing some whitepaper on this specific topic asap.
Sounds good if you want to create dataset but you'd need to pay for the GPU rent, exchange listing fees, and market making.
People like to be incentivized but if your altcoin price is too low nobody gonna use your gen AI and contribute and since your altcoin inflation seems to be on the higher side, be ready to get unlimited selling pressure.
It's at the end of the day, requires capital just like anything else to get started. I believe I also have seen some similar project but just can't seem to recall.
A good thing is, I almost have no "GPU rent" since I bought most of the infrastructure in past few years of running my company. Not huge GPU farms like what OpenAI has, but quite a bunch of NVIDIA, AMD and even newly made Chinese gears as infrastructure for making fine-tuned models and doing the inference. Honestly, most of them keeping us ahead like $4k/hour bills. Exchange listing and market making is an issue obviously.
Well, I was working on this idea with a friend who's residing in Istanbul and when we made it a little serious, my country shut the internet down. I had no access to the internet for around 90 days or if I had, it was with a very expensive VPN subscription.
The thing is, what we've done so far was like this: people generate photos on a small model (Flux Klein 4B and our own LoRA's on top) and give us feedback on the result. For validation of the result as good or bad - since they helped us identify which is which - we decided to reward them with $TON (or a TON based token).
It is good at its core but well... I wish for a more stable middle east to work on this type of globally interesting ideas.
You leave in the same country, Turkey with your friend, and why are they shutting down the Internet access, what's happening in Turkey to face this challenge, for 90 days? How you want people to communicate with clients abroad, that's ridiculous.
What's your gain when rewarding users with Ton coin, how do you generate your gain. Is it from the transaction fee or generating fee?
I am in Iran (next door neighbor of Turkey) and we had our internet shut down for 90 days due to the war. I lost access to my friend. He's still on the project but I'm losing my hope.
About the gain, I personally think about the affordable data gathering/validation. At least it's my current position.