Fraction AI: Decentralizing Data Labeling

There's a new player in town called Fraction AI, and they’re pretty ambitious. They want to disrupt the data labeling landscape. They just pulled in $6 million during their pre-seed funding round. The round was co-led by Spartan Group and Symbolic Capital, featuring some big names in the investment rungs. They were founded back in February, but it seems they're finally getting the traction they need. So, what’s the catch? Join me as I break down this new token launch and more.

Fraction AI has its eyes set on AI’s most crucial element: data. You may think that the models and compute power are most important, but without quality data, they’re nothing. The founder, Shashank Yadav, made it clear this is where they want to help. “Among AI's three core elements—data, compute, and models—data remains the most obscure and tightly controlled,” he said. “We set out to change that, leveling the playing field and empowering anyone to train high-quality AI models.”

The Appeal of Decentralized Data Labeling

Decentralizing data labeling has its advantages. By opening up the process to more sources, it can improve data diversity, reduce bias, and enhance data quality. It sounds great in theory, and it’s hard to argue with the logic. A decentralized system can offer data from many different perspectives, which can mitigate any biases that come from sourcing from a singular entity.

More than that, having a data marketplace that pools its resources can improve the overall quality of the data. If you’ve ever tried to label data yourself, you know how big a problem this is. You need clean, well-organized, and accurately labeled data to get good results.

The Mechanics of Fraction AI

Fraction AI is going the hybrid route that combines human insights and AI agents. There are three main participants in this process: stakers, builders, and judges.

The stakers stake ether or liquid staking tokens (LSTs), like Lido staked ether (stETH). They earn a reward pool funded by builders’ entry fees. That’s 5% of each entry fee, which is a nice return for doing nothing.

The builders create agents using human insights or detailed instructions. They get the process rolling by funding their agents with ETH or LSTs. To make a return, they pay a small entry fee to compete for the best data possible.

What’s interesting is that the top three agents from each of their competitions will be rewarded by the entry fee pool, but they will also earn multipliers based on how well they perform as rated by large language models (LLMs).

And the judges? Well, they’re the LLMs that evaluate the agents against some criteria. They’ll need to stake Fraction AI’s native FRAC tokens to participate.

Future Plans

Fraction AI has been raising funds since April and closed the round in September. The round was structured as a simple agreement for future equity (SAFE) with token warrants. It isn’t just the big names that joined this round. Some angel investors joined as well—think Sandeep Nailwal of Polygon and Illia Polosukhin of Protocol. So the project has some weight behind them.

Fraction AI is built on Ethereum and has been live on a closed testnet with 60,000 users. The public testnet is set to launch next month. The mainnet will be released by the end of the first quarter or early second quarter of 2025. The FRAC token will come out a little closer to the mainnet, and it’s supposed to secure the network of judges.

Blockchain’s Role in It All

Blockchain helps provide a secure and decentralized way to manage the data. That’s invaluable if the point is to have a decentralized cryptocurrency lab that is going to train AI models. There's an audit trail. The data is secure, and the data is trustworthy. That’s crucial for AI.

Also, blockchain and AI working together can speed things up as they automate tasks like data labeling, model training, and smart contract execution. It takes the human error element out of it.

Summary

Fraction AI is a promising venture that aims to decentralize data labeling. With its innovative hybrid approach and unique reward system, it could make high-quality data more accessible and reduce biases. As they gear up for their mainnet launch, I think it’s safe to say they could be a significant player in the crypto space. They'll need to be as the competition heats up.