AI has three basic components: a model, compute, and data. Add liquidity, so that smart agents can transfer value, and these four inputs are the basic building blocks needed to create smart agents. Let’s break down each of these parts and explain why each benefits from decentralization.

Models are algorithmic structures that process data to produce predictions, classifications, or other outputs. There are many different kinds, from simple linear regression models to complex deep learning networks such as convolutional neural networks. AI models, like any kind of software, can be open for public examination and used by all, or they can be closed and treated like an industry secret.

AI as it exists today is extremely secretive. Many of the most popular models today are closed-source and users have no way to evaluate them. Even open-source models often don’t reveal their model weights, training data, and other crucial features that affect their results. All of this makes it difficult for users to trust AI for more advanced tasks.

As we entrust more activity to AI, we need models that are transparent, verifiable, accountable, and censorship-resistant. Lack of transparency will create worse AI products. For instance, without visibility into the algorithms, it's hard to assess whether an AI model harbors biases. Models should be open source and accessible to all, and users should be able to verify a model’s training and output onchain.

The goal of Talus is to create actionable AI that can execute user intents. For this to happen, users need cryptographic assurances their intents were actually carried out. For AI to realize its potential, cryptographic trust must become the norm.

Data is food for AI models. The more high-quality data a model eats, the more it can grow and provide better results. The quality, quantity, and relevance of the data all affect the performance and capabilities of the AI.

In web2, most data is owned by just a few powerful web2 entities. This poses extreme risks to privacy, security, and accessibility and has caused many to look to web3 for its emphasis on digital ownership and privacy. Because so much data in web2 is controlled by just a few tech giants, AI is particularly vulnerable to manipulation and bias.

Web3 allows users to be the true owners of their digital assets. This can include users taking digital ownership of the data they create. If a company wants that data to train an AI model, users can receive compensation for their data or deny it all together. Likewise, startups and new AI projects that were previously at a data disadvantage can come together with users looking to earn from their data through data marketplaces. Creating infrastructure around data ownership will have benefits for AI and for all tech users.

Compute is the computing power required to train and run AI models. Compute power is crucial for processing the large volumes of data and complex algorithms used in training AI.

The hardware, software, and electricity needed to train AI are extremely expensive. This gives a massive advantage to existing web2 giants, who have these tools already or the means to build them rapidly. The result is continued concentration of control over AI in powerful, centralized institutions.

Decentralizing compute democratizes access to these hardware resources so that smaller teams like startups can also train and build AI. Just as decentralized data enables small developer teams to feed their models, decentralized compute enables these teams to conduct inference. It also creates a monetization opportunity for those with existing compute resources. Decentralized networks will emerge composed of numerous, specialized clusters of compute. The result will be cheaper, more accessible compute for all.

To put it bluntly, liquidity is cash. The concept of liquidity represents easy, instant, and cheap access to money. At Talus, we started with a simple vision of AI smart agents that can buy things for users. This would be the next huge unlock for AI’s capabilities–giving it the ability to handle money and transfer value. Nothing like this exists in web2, and we intend to make it possible on Talus.

A big reason why smart agents don’t exist in web2 is because AI cannot use traditional financial infrastructure. An AI smart agent can’t open a bank account, but it’s trivial to assign a wallet to an agent. In order to have a smart agent network, agents and liquidity need to live on the same interoperable and composable data substrate.

Permissionless value transfer is the bedrock of web3. It drives web3’s advantages, including trustless transactions controlled by smart contracts, censorship-resistance, manipulation-resistance, instant global transactions, freedom from external controls, hyperfast global consensus, and advanced social coordination. When it comes to smart agents, crypto is the best tool to use.

If AI is decentralized, its inputs are accessible to all, and its outputs are verifiable by all, the world will be a much better place.  Not only will decentralization lead to a more egalitarian AI future, it will allow us to create stronger, more performant AI products. Chief among these new products are AI smart agents, or AI that can transfer value and execute user intents autonomously. If we want smart agents, we need to apply principles of decentralization to models, data, compute, and liquidity. Talus is the nexus where these four elements come together to foster smart agents.

AI has three basic components: a model, compute, and data. Add liquidity, so that smart agents can transfer value, and these four inputs are the basic building blocks needed to create smart agents. Let’s break down each of these parts and explain why each benefits from decentralization.

Models are algorithmic structures that process data to produce predictions, classifications, or other outputs. There are many different kinds, from simple linear regression models to complex deep learning networks such as convolutional neural networks. AI models, like any kind of software, can be open for public examination and used by all, or they can be closed and treated like an industry secret.

AI as it exists today is extremely secretive. Many of the most popular models today are closed-source and users have no way to evaluate them. Even open-source models often don’t reveal their model weights, training data, and other crucial features that affect their results. All of this makes it difficult for users to trust AI for more advanced tasks.

As we entrust more activity to AI, we need models that are transparent, verifiable, accountable, and censorship-resistant. Lack of transparency will create worse AI products. For instance, without visibility into the algorithms, it's hard to assess whether an AI model harbors biases. Models should be open source and accessible to all, and users should be able to verify a model’s training and output onchain.

The goal of Talus is to create actionable AI that can execute user intents. For this to happen, users need cryptographic assurances their intents were actually carried out. For AI to realize its potential, cryptographic trust must become the norm.

Data is food for AI models. The more high-quality data a model eats, the more it can grow and provide better results. The quality, quantity, and relevance of the data all affect the performance and capabilities of the AI.

In web2, most data is owned by just a few powerful web2 entities. This poses extreme risks to privacy, security, and accessibility and has caused many to look to web3 for its emphasis on digital ownership and privacy. Because so much data in web2 is controlled by just a few tech giants, AI is particularly vulnerable to manipulation and bias.

Web3 allows users to be the true owners of their digital assets. This can include users taking digital ownership of the data they create. If a company wants that data to train an AI model, users can receive compensation for their data or deny it all together. Likewise, startups and new AI projects that were previously at a data disadvantage can come together with users looking to earn from their data through data marketplaces. Creating infrastructure around data ownership will have benefits for AI and for all tech users.

Compute is the computing power required to train and run AI models. Compute power is crucial for processing the large volumes of data and complex algorithms used in training AI.

The hardware, software, and electricity needed to train AI are extremely expensive. This gives a massive advantage to existing web2 giants, who have these tools already or the means to build them rapidly. The result is continued concentration of control over AI in powerful, centralized institutions.

Decentralizing compute democratizes access to these hardware resources so that smaller teams like startups can also train and build AI. Just as decentralized data enables small developer teams to feed their models, decentralized compute enables these teams to conduct inference. It also creates a monetization opportunity for those with existing compute resources. Decentralized networks will emerge composed of numerous, specialized clusters of compute. The result will be cheaper, more accessible compute for all.

To put it bluntly, liquidity is cash. The concept of liquidity represents easy, instant, and cheap access to money. At Talus, we started with a simple vision of AI smart agents that can buy things for users. This would be the next huge unlock for AI’s capabilities–giving it the ability to handle money and transfer value. Nothing like this exists in web2, and we intend to make it possible on Talus.

A big reason why smart agents don’t exist in web2 is because AI cannot use traditional financial infrastructure. An AI smart agent can’t open a bank account, but it’s trivial to assign a wallet to an agent. In order to have a smart agent network, agents and liquidity need to live on the same interoperable and composable data substrate.

Permissionless value transfer is the bedrock of web3. It drives web3’s advantages, including trustless transactions controlled by smart contracts, censorship-resistance, manipulation-resistance, instant global transactions, freedom from external controls, hyperfast global consensus, and advanced social coordination. When it comes to smart agents, crypto is the best tool to use.

If AI is decentralized, its inputs are accessible to all, and its outputs are verifiable by all, the world will be a much better place.  Not only will decentralization lead to a more egalitarian AI future, it will allow us to create stronger, more performant AI products. Chief among these new products are AI smart agents, or AI that can transfer value and execute user intents autonomously. If we want smart agents, we need to apply principles of decentralization to models, data, compute, and liquidity. Talus is the nexus where these four elements come together to foster smart agents.