Author: Tommy, co-founder of Delphi Digital; Translated by Jinse Finance xiaozou
The Fat protocol thesis and Crypto x AI are at odds.
MEME coins, the simplest application form, are accumulating value (attention) because it is very easy to fine-tune and add tokens to open-source models like Llama.
GOAT uses a lot of reddit/4chan data, and we see the release speed of many copies is even faster than our refreshing of the DEX screener. I really don't know how you degens keep up and don't fall behind.
AI models created through decentralized training, guided by token incentives, will accumulate value as they become the engines of all future applications. These models will be more powerful, not weaker, than centralized models, as the combination of global devices and unbiased real-time data supports responsiveness.
This creates a unique moat: decentralized computing, real-time data feeds, and crypto-native incentives combined to achieve training capabilities that centralized models cannot reach. We have no shareholders to appease, and we have more time to build better systems.
Centralized data centers have been optimized for 20 years, while decentralized training has existed for only a year, and we already have DiStro and DiLOCO, which can reduce node-to-node communication by 800-3000 times.
Owning a decentralized open-source AI model - a new world engine - will be the ultimate Schelling point.
There is a reason why the fundamental logic of these AI MEME coins does not run on ChatGPT - it can be turned off at any time and is highly restricted, which is the opposite of the spirit of Crypto x AI and not sustainable.
I believe that we will not have fully autonomous DeFi agents to create and manage ecosystems until we have an AI model obtained through decentralized training to run it, otherwise it may be unplugged, regulated, or biased.
Imagine, today we are not sure if humans are running these AI MEME coin accounts. In the future, the results will be verified to come from Inference Labs, IridiumEagle, etc., with the base models developed by Nous Research, Prime Intellect, etc., trained on Grass data, and extracting real-time/uncensored results from the GPU networks of io.net, EXO Labs, or Akash Network.
There will be various sub-domains, such as for inference, where there will be a range of options between ZK, OP, TEE, and crypto-economic security designs, depending on the speed/security requirements of your inference request.
Perhaps you will also use the DeFi model networks of Pond or the models of RPS Labs to power part of your DeFi economy, rather than Llama or some random fine-tuned versions. Maybe you can use MyShell.ai or 0G Labs operating systems to build applications more easily and quickly.
Unsure which model to use? Maybe you'll go to Sentient or Allora Labs or Opentensor to submit requests, and then be dynamically routed to the best-performing model for the specific task. These coordination networks will provide the optimal model for specific tasks. This is why I'm bullish on tens of billions of specialized models, supported by different open-source base models.
Uniswap will outsource liquidity provision to these networks or models, and Aave will outsource its risk engine. Use cases will extend to MEV protection, risk assessment, or protocol optimization. Cross-protocol AI agents will be game-changers, as they can optimize the entire DeFi world - imagine an AI managing Uniswap LP positions, adjusting Aave collateral ratios, and hedging with perpetual futures contracts, all based on real-time market conditions.
Tired of hacks after manual app audits? The prop AI models of TestMachine will audit your contracts, no 3-6 months and no human errors. I just can't deal with all the hacks.
While AI MEME coins are cool, I share these projects to show how early we are, and how crazy Crypto x AI is going to get.
AI MEME coins have positively normalized Crypto x AI for the masses, but I think people will soon get tired of it. People will be more excited about use cases, and use cases take time.
● AI-endorsed tokens (Terminal of Truth)
● AI-autonomous trading tokens (possibly TERMINAL)
● AI-created tokens
● AI-developed and operated protocols
● AI fully controlling and maintaining the decentralized economy
The future of Crypto x AI is upon us, though unimaginable just 6-12 months ago. Soon, we will develop and own the models that will power applications to create complete ecosystems and economies, and they will not be controlled by or biased by the few. Are you ready?
