laitimes

Deep Analytics AO: Bringing AI to Crypto's Supercomputer

author:MarsBit

原文标题:Research: AO, a F**King Big Computer

Original Author: Teng Yan

原文来源:chainofthought

编译:Luffy,Foresight News

Earlier this month, SpaceX successfully launched Starship for the fourth time, the most powerful rocket to date.

This engineering marvel boasts a super-heavy booster with 33 Raptor engines with a total thrust of 72 Bull. To put it another way, that's twice the thrust of the Apollo moon mission rocket and three times the thrust of the NASA space shuttle.

Deep Analytics AO: Bringing AI to Crypto's Supercomputer

I've always been fascinated by the development of tech companies. Years of unremitting efforts and huge R&D investment have finally brought about seemingly overnight success. But industry insiders know that for a high-tech company to succeed, there must be a myriad of low-probability events that fit perfectly. Founded 22 years ago as a startup with big dreams, SpaceX is the epitome of this journey – the result of hard work and continuous improvement.

Such a huge success is rare in the cryptocurrency space.

Of the crypto companies during the ICO boom of 2017-2018, only a handful still maintain their influence to this day. Cryptocurrency is a young and emerging industry with very short attention spans, and this mindset tends to influence founders' perception of the company.

That's why it's exciting to see the OG project grow to the next level.

Arweave is tightly connected to AO computers

Deep Analytics AO: Bringing AI to Crypto's Supercomputer

It took Arweave more than 5 years to break the 10 million transaction mark (November 16, 2023). Source: https ://viewblock.io/arweave/stat/tx

Launched in 2018, Arweave is a decentralized, permanent data storage protocol. While data storage is critical, it has struggled to find compelling use cases and adoption options.

Data storage isn't sexy, it's a commodity. The AO computer is Arweave's most ambitious project.

AR = Storage Layer

AO = Compute Layer

AO's goal is to be a scalable, secure, trustless, and customizable computing platform – hyper-parallel computers.

The core idea of AO is that deterministic virtual machine + permanent data store = reproducible state. This vision was in place four years ago, only now in AO.

I want to emphasize that AO computers are very different from most other blockchains today. It exists thanks to the groundwork that Sam (the founder) and the Arweave team have done over the years. Just as high-tech startups need a solid foundation to succeed, AO wouldn't be possible without Arweave's secure decentralized storage network.

Not another Ethereum or Solana

Deep Analytics AO: Bringing AI to Crypto's Supercomputer

Ethereum and Solana are global synchronous state machines with shared global state, while AO machines are shared global messaging machines with local state.

Sharing global state is the default mode for today's blockchains. It enables participants to reach consensus and trust, thus maintaining the integrity of the ledger. However, this is not enough to support future AI applications.

The AO computer framework differentiates itself from Ethereum and Solana through two main features: (1) scalable computing power and (2) a flexible security model.

Scalable computing

In Ethereum and Solana, adding more compute resources (new nodes) does not increase the throughput of the network. No matter how many validator nodes there are, Ethereum will still only process 12 - 15 transactions per second.

However, the AO computer framework provides a dynamic way to scale. As more computing resources are integrated into the network, throughput scales accordingly. This is because AO processes run locally in parallel and are not limited by global state.

Security model

Ethereum has a unified security model where every transaction is equally protected in its proof-of-stake (PoS) network. While this ensures consistency, it can be inefficient and costly for low-value transaction scenarios such as gaming.

The AO computer framework introduces a flexible security model that allows developers to tailor security requirements to the specific needs of their processes.

supercomputer

Imagine you have a giant supercomputer made up of many small computers all over the world.

This large networked computer can perform multiple tasks at the same time, such as running games and applications. Small computers communicate with each other using a messaging system. The advantage of this is that each small computer is not slowed down by the rest of the network, but still gains the security of the blockchain.

That's what AO computers are, in a nutshell.

The AO computer framework upgrades the typical blockchain node architecture by breaking it down into smaller, modular components. These components include:

  • process
  • Messaging unit
  • Dispatch unit
  • Compute Units

Each element plays a key role in the overall functionality of the system (complete details are provided in the documentation and white paper).

Deep Analytics AO: Bringing AI to Crypto's Supercomputer

Source: Messari

At a high level, AO can be boiled down to three core concepts:

  • Multiple processes that can run in parallel locally
  • The process preserves individual sovereignty
  • Asynchronous messaging, which allows coordination and communication between processes

Parallel processing

At the heart of the AO framework are processes, i.e., applications on the platform. These processes run independently and in parallel locally. Each process has no memory and can't see the local state of other processes.

This isolation allows for complex calculations on powerful machines, expanding the design space far beyond the limits imposed by traditional smart contracts (fees, gas limits, block sizes).

The computing unit plays a key role here. They provide the necessary computing power to keep the AO process running efficiently.

Apply sovereignty

A distinguishing feature of AO is that it gives developers sovereignty.

Each process runs autonomously, allowing developers to choose their computing tasks, virtual machines, and security parameters without interference from other processes.

Developers can adjust security measures based on the value and sensitivity of a particular computation.

Asynchronous messaging

This is the glue that holds the network together. Communication within the AO protocol is managed through asynchronous messaging.

Processes exchange messages to coordinate operations and receive the necessary data. All applications follow common messaging standards, ensuring consistency and interoperability.

The messaging unit delivers these messages, working in tandem with the scheduling unit (similar to a rollup sequencer) to sort the messages and record them on Arweave.

The trillion-dollar business opportunity for AO lies in AI

It's not realistic to run a large AI model with billions of parameters like LLama-3 directly on Ethereum.

  • Compute requirements, data storage, and bandwidth: The LLaMA-3 model requires significant computing resources (RAM, GPU capabilities). Ethereum is not optimized for the large-scale data storage and high bandwidth required by LLMs.
  • Latency and performance: Ethereum's transaction speed struggles to meet the low latency, high throughput required for AI models. No user wants to wait 12 seconds (Ethereum's block time) for a result.
  • Cost: On-chain AI computing is very expensive.

So I was very excited when AO announced last week that large language models (LLMs) could run on smart contracts, bringing AI directly to the blockchain. This will integrate human-like decision-making in a trustless network.

memory

Running an AI model requires loading model parameters into memory, and the more parameters there are, the greater the memory requirements.

Running a relatively small model, such as LLama-3-8B, requires at least 12GB of RAM. GPT-4 has more than 1.76 trillion parameters, and GPT-5 is expected to have 50 trillion parameters. What's even more terrifying is that AI models are only going to get bigger.

Deep Analytics AO: Bringing AI to Crypto's Supercomputer

Source: AO Computer

Current blockchains are very limited in terms of memory. At the protocol level:

  • Ethereum has 48KB of RAM when it comes to executing smart contracts
  • Solana 有 10MB 的 RAM
  • 即使是正在进军人工智能领域的互联网计算机 Internet Computer (ICP),其 RAM 也只有约 3B。

AO uses WebAssembly in its execution environment and now supports WASM64 to support high-performance applications. Currently, AO has a memory limit of 16GB and a protocol-level cap of 18 exabytes. This allows it to run heavy calculations, such as inference on Llama-3 or Phi 3.

Data storage

WeaveDrive is a new feature that enables AO applications to access all of Arweave's data as if they were a local hard drive.

By acting as a local file system in a smart contract, WeaveDrive improves the efficiency and accessibility of data storage, retrieval, and retrieval. More DApps will be incentivized to upload and store data on Arweave.

AO Computer's framework leverages the computing power of a single machine to enable direct on-chain execution of large AI models.

By combining an unlimited execution environment with efficient and accessible data storage, AO significantly expands the design space for on-chain applications, especially those that are computationally intensive and AI-driven. This facilitates the development of autonomous agents with similar smart contract guarantees.

Integrating cron jobs (a feature that most blockchains don't have on their own) and autonomous proxies will significantly boost on-chain activity, facilitating more complex and dynamic interactions.

Note: Cron is a computer term that denotes the execution of work that has been scheduled at an agreed time.

As we move toward an era where AI is ubiquitous, many would underestimate the importance of this breakthrough.

Deep Analytics AO: Bringing AI to Crypto's Supercomputer

Llamaland as a proof-of-concept example: an AI-powered MMO game built on AO computers.

At Llamaland, users can submit petitions to Llama King, an AI agent that uses a large language model to review users' petitions and decide how much Memecoin to allocate to each user. It's worth noting that the entire process is executed entirely on-chain, which is the first implementation of an LLM that runs entirely on-chain.

Although it's not live yet, you can sign up as an alternate user.

Build a permanent network

Deep Analytics AO: Bringing AI to Crypto's Supercomputer

Odysee is a video-sharing platform

On June 6th, Arweave made 3 important announcements aimed at expanding the market segment:

  • Forward Research 收购 Odysee
  • Autonomous Finance 推出 AgentFi
  • AO Ventures is providing $35 million in funding to builders of the AO ecosystem

From my point of view, the most striking was the acquisition of Odysee. Most people underestimate its value, and Odysee is the biggest Web3 social app that few people have heard of. It has more than 7 million monthly active users (20 times more than Farcaster) and 2.2 million monthly website visits.

Deep Analytics AO: Bringing AI to Crypto's Supercomputer

来源:Similarweb

Odysee is an open version of YouTube that leverages blockchain technology to give creators greater control and freedom. Unlike YouTube, which relies on traditional ad revenue, Odysee rewards creators with LBRY Points (LBC).

Odysee is more like Reddit, a dark corner of the internet that emphasizes free speech with less content filtering. Its "Wild West" classification is particularly notable for hosting more controversial and unfiltered content.

SimilarWeb says a significant portion of its users are from Europe, particularly France, Spain and Germany. Even though Odysee can't match YouTube's scale, it still attracts a large number of users who may be potential users of AO's app.

Odysee will now build on Arweave infrastructure, adding its platform to the permanent network and giving creators full control over their content.

AO

Deep Analytics AO: Bringing AI to Crypto's Supercomputer

Source: ao.link, data as of June 19, 2024

AO has now released a testnet with proof-of-authority (PoA) security. Most applications on mainnet will employ a proof-of-stake (PoS) mechanism. The mainnet launch date has not yet been determined.

Over the past three months, the testnet has had about 7,000 daily active users, 3,000 to 4,000 processes, and 182 million messages sent. While these numbers are not staggering, they indicate the existence of a small but active community.

Dexi

Deep Analytics AO: Bringing AI to Crypto's Supercomputer

Dexi automatically collects and manages data on asset prices, swaps, liquidity, and more. Think of Dexi as a blockchain-based Bloomberg. It is managed by a network of autonomous agents, hosted on Arweave, connected to all available liquidity pools.

Dexi is permissionless, and users can pull data directly from the Dexi aggregation agent. It is censorship-resistant, and all information is verifiable on-chain.

0rbit

Deep Analytics AO: Bringing AI to Crypto's Supercomputer

0rbit is a decentralized oracle network with two main components: an AO module and a network of nodes that fetch data from anywhere on the internet. 0rbit is the critical infrastructure for building applications on AO.

The AO module interacts with the 0rbit node to send data requests and receive results. Anyone can run a node and contribute to the network. The data is fetched asynchronously from the internet, which means that the process doesn't have to wait for a response.

Currently, 0rbit can fetch news from a user-defined website API and price information from Coingecko's API.

As I fumbled, I discovered some other interesting things on AO, including AMM (Bark, Permaswap, ArSwap), Trunk (an AO meme coin), and Astro (an overcollateralized stablecoin that uses AR as collateral). AO is still in its early stages, and many projects have not yet been tested in production or amassed with a large number of deals.

Focus: AO tokenomics

Deep Analytics AO: Bringing AI to Crypto's Supercomputer

The AO token went live on June 14, with a minting date retroactive to February 27, 2024. I like its fair issuance model and no dedicated allocation of shares for investors and teams.

The AO token minting program is similar to Bitcoin, with no pre-mining and a total of 21 million. The rate at which new tokens are minted halves every 4 years, and unlike Bitcoin, this is a smooth, slow process of reducing production.

The new AO tokens will be distributed as follows:

  • Arweave Token Holders (36%)
  • stETH、SOL 等资产质押者(64%)

So far, around 1 million tokens have been minted, most of which have been distributed retroactively to AR token holders. AR token holders will continue to receive new AO tokens every 5 minutes. The testnet token AOCRED can be exchanged for AO at a 1000:1 ratio.

Important: AO tokens are not transferable or tradable until 15% of the total supply has been minted (approximately February 8, 2025). This provides time for the development of the emerging ecosystem and the launch of the mainnet, and the AO token will be used to secure the network.

Hold AR to earn AO

Below I have done some calculations on the expected return of holding AR tokens, here assuming 1 AR = 30 USD.

  • Holding 1 AR will receive about 0.016 AO tokens in the next 12 months.
  • There will be about 4.2 million AO in circulation over a 12-month period

The APY for holding AR is modest, ranging from 2.5% to 50%, depending on the expected FDV, which in turn depends on market conditions when the token becomes tradable next year. Assuming good market sentiment in 2025, a $10 billion FDV could be quite reasonable, implying an APY of 25%.

Earn AO by staking assets (e.g. stETH).

Deep Analytics AO: Bringing AI to Crypto's Supercomputer

Users who hold 0.01% of the total value locked (TVL) will accumulate about 210 AO tokens over 12 months.

As of June 25, the value of cross-chain stETH is around $320 million, and according to TVL and FDV, the APY is very impressive, at 33% - 600%. Estimated APY is volatile and heavily influenced by the total value of cross-chain assets (TVL) and the expected token FDV.

This is a very attractive low-risk yield opportunity for anyone holding stETH today.

As more users become aware of this and stake their stETH, the APY will drop. The idea of stETH mining is a win-win situation for the AO team and its users.

The staked stETH generates a significant amount of revenue ($1 billion in deposits, or about $30 million per year), which will be used to fund ecosystem projects and other development activities.

Deep Analytics AO: Bringing AI to Crypto's Supercomputer

Source: Dune@zkayape

Keeping an eye on the TVL of the staked asset is crucial, as it may eventually balance out with the AR yield. If TVL remains relatively low, more people may sell AR to mine using staked assets and vice versa.

The total supply of AR is 66 million, of which 65 million are in circulation. AR has a potentially deflationary characteristic, as the AR used for data storage goes into an endowment fund, which releases AR tokens only when mining is unprofitable.

My opinion on AO

When considering whether a protocol will be successful, I like to break it down into smaller questions.

I asked myself: What are the most important things AO needs to do well to be as successful as Solana?

  • There is a high demand for AI agents and applications in the cryptocurrency space: This seems to be the current trend given the rapid development of AI over the past two years. Other blockchains are developing off-chain computation using technologies such as zkML/OpML, but currently only AO supports full AI computation directly on-chain.
  • Composability with major blockchains: AO must be tightly integrated with other major blockchains. It's unrealistic to expect existing developers and applications to fully migrate to AO. Therefore, it is essential to build bridges and communication channels with existing ecosystems.
  • Develop a social layer: At the end of the day, blockchain success doesn't lie in transaction speed or low fees. Cultivating a passionate developer community and engaging users is key to widespread adoption.

I will pay close attention to these three areas and adjust my confidence accordingly.

Some other thoughts

  • AO computers handle decentralized computing in a very different way than traditional blockchains. This can be a success or a failure. But the ability to perform large computations on-chain is a key in the era of artificial intelligence.
  • AO takes a participant-oriented approach. These platforms are used in systems that require high concurrency, modularity, and scalability. While the actor model has theoretical advantages, its widespread adoption and large-scale success have yet to be demonstrated. Challenges include dealing with unexpected failures and maintaining consistency among decentralized participants.
  • AO is still in its very early stages. When testing the online experience, I found that users were mostly experimenting with text chat and gaming. AO needs to build tools, and that takes time. Due to the different architectures and languages used, it is not easy to port applications directly from other chains. Overall, AO feels like Solana in early 2021, when it only had a handful of apps and more skeptics.
  • Long-term belief is required: AO requires a high degree of conviction. Short-term traders may struggle to find an instant catalyst as the coin won't be tradable until 2025.
  • Arweave demand: AO will create an ongoing demand for AR as both a growth catalyst and a new platform for application development. Every process on AO will write data on Arweave and drive the usefulness of AR.

Artificial intelligence and cryptocurrencies are undoubtedly two of the most important technological paradigm shifts of this century. The intersection of artificial intelligence and cryptocurrencies will be transformative. My gut told me that AO was going to be a key project to lead the way.

Read on