Abstract

Decentralized AI (DeAI) is a growing narrative of interest for crypto investors but most of the projects with real technical innovation in this category are yet to emerge. This sub-category of the crypto industry has yet to develop and incorporate the needed infrastructure to support the training of AI models as well as operating the models with live-human interactions. Additionally, the technology is not magic and there are computer science tradeoffs, strengths and weaknesses for every technological choice that are critical to understand given a specific task. AI models built to process binary files such as audio, images, and video require a very different tech stack than applications designed to process time series data, as well as processing text.

Introduction

The crypto narrative of Decentralized Artificial Intelligence (DeAI) is one of the most bullish niches with high potential growth for the 2024/2025 bull market. However, there are a large amount of misconceptions regarding this industry because of the lack of understanding of the technology. Any programmer can develop websites and simple applications, however it requires a much more significant amount of knowledge to conceive and implement real innovations when it comes to artificial intelligence simply because of the complex mathematics involved in powering the various algorithms. Often, this is accomplished by research scientists with PhD’s working in some of the top public and private research centers in AI. Most web3 startups do not have access to such scientists on their team, and have ill-conceived understanding of the technology and are simply slapping a keyword on their website trying to fool investors into buying their token. As well, it can often take years of research and development before the technology works at a commercial level, ChatGPT, for example, took OpenAI more than 5 years and hundreds of millions of dollars before it became commercially viable. Decentralized AI requires an additional level of complexity and expertise, so let me go over the few projects I have identified that are really bringing innovation to the DeAI industry.

DeAI is the new GameFi

Last bull cycle, one particular crypto narrative generated massive enthusiasm and grew at extreme speed making many new millionaires within just a few months… This narrative was play-to-earn, a.k.a. GameFi. Axie Infinity grew in their first crypto cycle $1130X from $0.1457 (Nov 8, 2020) to $164.90 (Nov 8, 2021). This means a $1000 investment was worth $1.1M a year later. The Sandbox grew 184X from $0.04535 (Nov 25, 2020) to $8.40 (Nov 25, 2021). Thus, a $1000 investment was then worth $184,000 a year later. Decentraland grew 64X from $0.09 (Nov 25, 2020) to $5.85 (Nov 25, 2021), i.e. a $1000 investment was then worth $64,000. Overall, this means picking a few good projects within a trending narrative in conjunction with a bull market can lead to converting a few thousand dollars into a few millions within a single year. If DeAI proves to be one of the hot narratives in the crypto cycle, we might just be lucky enough to pick some of the winning projects. Note, that the fully diluted market cap of Axie, Sandbox, and Decentraland in Nov 2020 was just around $39M, $136M, and $197M respectively.

Any legitimate and innovative DeAI projects with a fully diluted market cap below $100M and direct relevance to consumers is likely to be a winning proposition. However, I do not offer investment advice, please do your own research, and understand that any of my favorite projects could lead me to a complete loss of capital. My thoughts are only described to express what are my own views regarding the industry, after all I do have 2 degrees in AI and spent nearly a decade in that industry prior to migrating to crypto. For most of the projects I will talk about, I have made investments prior to this blog post because I do believe in them, but I could be completely wrong for various reasons. Even if I’m not wrong, hackers, and even regulation could successfully destroy any of these projects.

DeAI Landscape

First, let’s describe a simplified data pipeline and process in bringing AI models to commercial viability: 1) defining the goal and selecting an AI model for the task, 2) data collection, 3) data processing (labeling, cleaning, etc), 4) training the model, 5) testing & refinements, 6) optimizing the model for operational efficiency, 7) deployment and live interactions with users, 8) additional data collection with user interactions and periodic retraining after eliminating data bias, etc. Each of these steps may require different tech stacks optimized for the task at the end, running everything on a decentralized architecture is likely not going to be viable at the current time point, which means there are plenty of commercial opportunities for web3/AI entrepreneurs.
In the infographics at the top of this post, I categorize the existing technologies I deem viable for specific types of tasks. The AI models that are commercially viable vastly differ for tasks optimized for processing text, binary files (images, video, documents), and time series data (such as a price feed of your favorite token). This is true for the data storage, to the compute layer, to the application layer. In order to run the entire data pipeline on a decentralized architecture for both training and operational interaction with users we would have to pick the right tool/technology at every layer, it’s not a one size fits all.
In the web2 world, computer scientists understand the infrastructure requirements for various applications. For example, graph databases (Neo4J) differ vastly in computational efficiency for certain tasks versus sequential databases (MySQL) or for document management (MongoDB). The right database can lead to significant cost savings and real time user interactions or sluggish performance.

Decentralized Storage Layer

There are significant misconceptions by many very influential crypto investors regarding Filecoin when it comes to supporting decentralized AI. Unfortunately, Filecoin has very low cost of storage and the tradeoffs for this was that it is optimized for cold/long term storage of data, i.e. not ideal for training or operating AI models. AI models during the training phase require a low latency database with large data throughput for bulk transfer meaning we need a hot storage mechanism. Depending on the tasks, I identified Jackal that uses a bucket storage mechanism for hot storage at higher cost than Filecoin’s cold storage mechanism, but lower cost than the alternative AI-optimized storage on AWS. Therefore, I would categorize Jackal as a good decentralized AI storage protocol for binaries and documents.

Decentralized Compute Layer

Once you have your data storage handled, you would deploy and train an AI model on a compute farm (GPU). Most cloud service providers such as AWS and Google provide such AI optimized cloud compute service. However, in the case of DeAI, you could train a model on Render, but it wouldn’t be ideal when you are trying to operate your AI model with users, for that reason, I believe BitTensor is vastly better than Render as a compute protocol for the DeAI industry, while Render is perfectly acceptable for the training phase, and later deployment on a centralized cloud provider for servicing end-users. BitTensor was designed to run various AI models such as Transformers (text processing, etc), Convolutional Neural Networks (images & video processing), Recurrent Neural Networks (time series, etc), and Reinforcement Learning (for example playing games at a level humans cannot compete), Generative Model, etc. However, it is entirely possible that other compute protocols could be better optimized for certain tasks which appear to be the orientation that Allora chose in regards to processing time series data, for example making predictions for trading and other financial applications.

Decentralized Application Layer

BitTensor has been used to develop and deploy Corcel, a multi-purpose AI application that has a: 1) Chat, 2) Image Generation, 3) Image Processing, 4) Image Search, 5) translation service. Each of these are built on different AI algorithms. Corcel Studio for example is a competitor to the centralized AI service MidJourney, which has made some considerable breakthroughs using a stable diffusion algorithm for generating high quality images based on textual inputs.


Checkout the latest video generation tool Sora from OpenAI, it is mind blowing to think what this will be in 5 years time.

Meanwhile Corcel Chat is a competitor to ChatGPT from OpenAI. I would tend to think that chatbots such as Corcel Chat or ChatGPT would be ideally suited for using graph databases such as Neo4J, however I’m not aware of any such graph decentralized storage outside of The Graph which was not designed as a storage protocol but as an indexing service.
Weavechain is another project about to launch a token that has real capability in terms of Natural Language Processing. They have a variety of upcoming products, but their first one was designed to identify copyright infringement using Large Language Models. Their team is highly focused on building DeAI technologies with real commercial viability instead of pie in the sky magical thinking.

Summary

Given that the Decentralized AI industry is in its very nascent stage, I believe this represents excellent business opportunities for both entrepreneurs and investors to support the high growth potential of the industry. Given the computer science challenge and very diverse set of AI algorithms and tasks, there is room for many players to gain market share. I would highly welcome any recommendations of projects to add to this landscape and infographic.

About

Crypto Rookies is a crypto investor, serial entrepreneur in Artificial Intelligence and Web3/crypto with expertise in tokenomics and market making. Currently CEO of Smooth, which focuses on solving the problem of 95% of crypto-currencies failing in their first 2 years.
Smooth prevent such failure by:
1) Community building with AI (don’t waste money on marketing & influencers)
2) SEC-compliant Tokenomics
3) Market Making (hybrid between stablecoins & altcoins)
4) Treasury management (10%-100% APY depending on risk):
This blog channel generates 10,000+ views per month on the topic of tokenomics and crypto investments, so if you are interested in becoming a sponsor, please reach out to Crypto Rookies.
Meanwhile, please feel free to join the various communities I am involved in and do not hesitate to reach out.

1. Twitter
2. YouTube with Cyrator and Mikhail Yergen
3. Cyrator (a transparent and reliable token review/ranking community where anyone can join, contribute and earn)

Also check out some of the early stage crypto assets that I’m actively using:

1. GRVT, it’s a Decentralized zk-powered crypto exchange.
2. DOP, it’s a privacy focused tool to hide crypto assets & transactions in your digital wallet.
3. MarginFi, it’s a peer-to-peer lending protocol on Solana.
4. Drift, it’s a Decentralized crypto exchange on Solana.
5. Swell, it’s a liquid staking protocol to use in conjunction with EigenLayer.

Use my referral code below for various crypto tools used for trading:

• TradingView for charting

Centralised Crypto Exchanges to trade with Leverage and Grid Bots:

• OKX
• BingX
• MEXC
• CoinW
• ByBit
• BitGet