Author: Tommy

In the combination of crypto and AI, the traditional fat protocol theory has been overturned.

Memecoins are the simplest form of application, and they are accumulating value (i.e., attracting attention) because it is easy to fine-tune open-source models like Llama and add tokens.

GOAT leverages a large amount of data from Reddit and 4chan, and we see many imitators emerging faster than we can refresh the DEX screener. I really don't know how these speculators keep up.

I believe this situation will change in the future.

Web2 will stop sharing new foundational models and will increasingly impose biases and restrictions; people will also grow tired of basic large language models and will crave truly integrated AI applications with crypto.

Currently, Llama is the preferred model for all AI memecoin developers, so we may be affected by Zuckerberg's restrictions on future version releases. Goodness, if Llama's users exceed 700 million, you'll need permission from Meta. Do you think he will agree?

AI models created through decentralized training, incentivized by tokens, will accumulate value as they become the core engine of all future applications. These models will be more powerful than centralized ones, not weaker, because we will combine global devices with unbiased real-time data to drive responses.

This creates a unique moat: the combination of decentralized computing, real-time data streams, and crypto-native incentives makes our training capabilities unmatched by centralized models. We have no shareholders to please, allowing for a longer timeline to build better systems.

Centralized data centers have 20 years of optimization experience, while decentralized training has just begun a year ago; we already have DiStro and DiLOCO technologies that have reduced communication between nodes by 800 to 3000 times.

Owning a part of decentralized open-source AI models, this new world engine, will become the ultimate focal point.

There is a reason why the fundamental logic of these AI memecoins is not running on ChatGPT. ChatGPT could be shut down at any time, and there are many restrictions, which is completely contrary to the idea of crypto x AI and cannot exist sustainably.

I believe that only when we obtain an AI model through decentralized training to run it can we achieve fully autonomous DeFi agents that create and manage ecosystems. Otherwise, it may be shut down, regulated, or biased.

Think about it, today we aren't even sure if these AI memecoin accounts are run by humans. In the future, you will be able to obtain results through verification reasoning (like @inference_labs and @IridiumEagle), which rely on foundational models trained using data provided by @getgrass_io (like @NousResearch and @PrimeIntellect), and obtain real-time and uncensored results through the GPU networks of @ionet, @exolabs (clusters), or @akashnet_.

These fields are becoming increasingly diverse. In the realm of reasoning, depending on your speed and security needs for reasoning requests, ZK, OP, TEE, and cryptoeconomic security designs will form their respective subfields.

You might choose to use the DeFi model network provided by @PondGNN or specific models from @RPS_Labs to support your DeFi economy, instead of relying on Llama or other randomly fine-tuned versions. You may also use @myshell_ai to build applications more easily and quickly, or implement them through the operating systems of @0G_labs.

If you're unsure which model to use, you can request @sentient_agi, @AlloraLabsHQ, or @opentensor, which will dynamically guide you to the best-performing model for your task. These coordination networks are entirely capability-based systems that can showcase the best models for specific needs. Therefore, I am confident in the billions of specific models driven by diverse open-source foundational models.

Uniswap will outsource liquidity provision to these networks or models, while Aave will outsource its risk engine. Application scenarios will expand to include MEV protection, risk assessment, and protocol optimization. The real game changer will be cross-protocol AI agents that can optimize the entire DeFi ecosystem—imagine an AI that can simultaneously manage Uniswap's liquidity pool positions, adjust Aave's collateral ratios, and hedge using perpetual futures, all based on real-time market conditions.

Tired of manual audits and still encountering application hacks?

@testmachine_ai's prop AI model can audit your contracts without waiting 3-6 months and avoiding human errors. I really can't take any more hacks. While AI memecoins are fun, I share these projects to show that we are still in the early stages, and how crazy crypto x AI is going to get.

Here is the crypto x AI flowchart I drew in May 2024:

AI memecoins have gradually made the public accept crypto x AI, but I think people will soon grow tired of quick forks. Everyone will expect actual application scenarios to emerge, which takes time.

1 -> AI-supported Token (the truth terminal has arrived)

2 -> AI auto-trading Token (perhaps the terminal)

3 -> AI creates Token

4 -> AI builds and operates protocols

5 -> AI fully controls and maintains decentralized economy

While this was taboo 6-12 months ago, a complete crypto x AI future where we build and control the models ourselves is now within reach. These models will drive the development of applications, creating a complete ecosystem and economy that is no longer controlled or biased by a handful of people in San Francisco. Get ready for this future.

To learn more, you can check out the crypto x AI monthly video content from @Delphi_Digital, supported by @autonolas.

Disclosure: Many exciting @Delphi_Ventures portfolio companies mentioned in the article welcome collaboration with them!

I may hold the tokens mentioned in the article. There are many reasons why this theory might fail or change, and I will write another post to discuss it. I notice that AI memecoins are extremely risky, and you may lose all your principal in most cases.