AI agents had a meteoric rise - and an even faster fall However despite market sentiment, we are just at the tip of the AI x Crypto iceberg. What is DeAI and how will it create a paradigm shift in the way we think of AI development?

Think about your interactions with AI today. When you use ChatGPT, where does all that data go? Where is it stored? Who can access it? But then looking even deeper - how are these models trained? Where is the training data coming from and is it legit? You have no way of answering any of these questions because the current state of AI relies on centralized, closed source systems that don’t let you peek behind the curtain.

As we grow more reliant on the use of AI, it’s important to think about these issues. If you rely on AI for accurate responses, analysis, and guidance, you’d want that model to be trained on data that reflects these values. When it comes to AI training, the thing you have to remember is ‘garbage in, garbage out’ - a model is only as good as the data it is trained on. With closed source models, how can we ensure that this is not the case? How do we know the answers they’re generating are using peer reviewed professional studies and not Wikipedia pages? The answer is simple: we don’t.

Decentralized AI (DeAI) seeks to address these exact issues by opening up the entire process of training an AI model from start to finish. This includes: - Collecting training data - Providing compute for training - Analyzing training algorithms and detecting bias - Verifying the accuracy of each result How is this all being done on blockchain today?

Let’s start from the beginning: data collection through @touch grass What is Grass? It’s a decentralized protocol who’s goal is to ethically scrape data for use in training AI while rewarding its operators with crypto. Companies are already utilizing your bandwidth to monitor you - without compensating you for said provided bandwidth. Grass uses the same technique to scrape data from public sites, but offers you an incentive to use their system while also preserving your privacy, unlike big tech. How does it work?

@touch grass runs in the background of your browser and scrapes info from the sites you visit without ever interacting with your personal data. Based on your contribution amount, you get rewarded in Grass points, which convert into their $GRASS token. Scraped data is then cleaned and structured to be used for training AI. Most importantly though, each piece of data is tagged with metadata so that it can be directly traced to it’s origin. The metadata contains the source URL, the timestamp of when it was scraped, and the node that conducted the scraping. All of this data is verified using ZK proofs to prove that it is accurate. All the data is then stored within the Grass Data Ledger built on Solana, where anyone can prove the origin of each dataset using the attached metadata.

@touch grass decentralizes the process of scraping data through incentive mechanisms, making it insanely efficient at gathering raw data for use within AI. In Q1 of 2025 alone, they scraped 57 million (!!) GB of data using this model. Check out the Grass dashboard here to see their current metrics: https://www.grassfoundation.io/network/stats

While scraping data for training is essential to DeAI, so is actually training these models in a decentralized fashion. This is where protocols like @The Bittensor Hub come in. Bittensor aims to democratize the entire process of training AI models through subnets - each subnet specializes in a different portion of the training process. How do these subnets work and what does the Bittensor incentive structure look like?

Subnets are operated as individual entities within the Bittensor ecosystem where a team runs one or multiple subnets with each one working on a specific aspect of AI modeling. For example, @Macrocosmos runs: SN1 - natural language and inference training SN9 - creates open source pre-trained models for testing SN13 - operates data scraping/storage SN25 - engages in protein-folding research SN37 - participates in fine-tuning finished models

The entire @The Bittensor Hub ecosystem operates financially through the use of the $TAO token. SN0 is the Root subnet responsible for the delegation of $TAO to individual subnets, where delegation is determined by a group of 64 validators that assess the work of each subnet to determine the allocation amount. Each other subnet essentially acts as a data miner, where their mining capabilities and effectiveness determine the amount of $TAO the subnet receives. External participants not directly working with Bittensor can also stake their $TAO tokens, which are added to the allocations, in exchange for yield.

The beauty of DeAI is that at scale, it’s able to work much more efficiently than centralized AI companies like OpenAI because it encourages everyday people to contribute to the process using crypto as an incentive mechanism. Want to earn $GRASS? Download the Grass browser extension and passively contribute bandwidth to the project. Want to earn $TAO? Stake your crypto within Bittensor and passively contribute to funding subnets. These mutually beneficial systems are what sets crypto apart from traditional industries - big tech will use your data without ever compensating you for it, and crypto turns the tables to let small companies utilize their communities to grow.

Decentralized AI is promising a bright future. Don’t let shitty markets ruin your mentality Enjoyed this thread and want to learn more? Check out my free blockchain course at http://dcft.site to become a blockchain pro today!

Some great additional reads if you found this content insightful: List of all Bittensor subnets - https://learnbittensor.org/subnets The real value in Web3 AI - https://x.com/Defi0xJeff/status/1910634579654041891… Where top VCs think crypto x AI is headed to next - https://www.coindesk.com/business/2025/04/12/where-top-vcs-think-crypto-x-ai-is-headed-next… Discover Grass - https://www.getgrass.io/learn