Age of Intelligent Beings: The Convergence and Symbiosis of AI and Crypto

Original Article Title: "The Era of Intelligent Agents: The Confluence and Symbiosis of AI and Crypto"

Original Author: YBB Capital Research

1. Originating from Attention's Love of novelty

Over the past year, due to a narrative breakdown at the application layer unable to match the speed of infrastructure proliferation, the crypto space has gradually turned into a game of capturing attention resources. From Silly Dragon to Goat, from Pump.fun to Clanker, the love of novelty in attention has made this competition increasingly insular. Starting with the most cliché eyeball-attracting monetization, swiftly transitioning to a platform model where attention seekers and suppliers are unified, and then to silicon-based organisms becoming the new content suppliers. Within the bizarre carriers of Meme Coins, a presence has finally emerged that can bring retail investors and VCs to a consensus: the AI Agent.

Attention is ultimately a zero-sum game, but speculation can indeed drive brutal growth. In our article about UNI, we once revisited the dawn of a golden age on the blockchain. The rapid growth of DeFi originated from the era of LP mining initiated by Compound Finance, where various mining pools with APY in the thousands or even tens of thousands saw the most primitive form of on-chain gameplay with constant entries and exits, although the eventual result was the collapse of various mining pools into chaos. However, the frenzy of gold miners did leave unprecedented liquidity to the blockchain, and DeFi eventually transcended pure speculation to form a mature track, meeting users' financial needs in various aspects such as payment, trading, arbitrage, and staking. At the current stage, AI Agents are also experiencing this brutal phase. What we are exploring is how Crypto can better integrate AI and ultimately propel the application layer to new heights.

2. How Intelligent Agents Become Autonomous

In a previous article, we briefly introduced the origin of AI Meme: Truth Terminal, and the outlook for AI Agents in the future. This article focuses firstly on AI Agents themselves.

Let's start with the definition of an AI Agent. Agent is a rather old but vaguely defined term in the field of AI, mainly emphasizing autonomy, that is, any AI that can perceive the environment and make reflexive decisions can be called an Agent. In today's definition, an AI Agent is closer to an intelligent agent, setting up a system to mimic human decisions for large models. In the academic world, this system is seen as the most promising path to AGI (Artificial General Intelligence).

In the early versions of GPT, we could clearly perceive that large models were very human-like, but when asked many complex questions, these large models could only provide somewhat ambiguous answers. The fundamental reason was that the large models at that time were based on probability rather than causality. Additionally, they lacked human abilities such as the use of tools, memory, and planning. An AI Agent can address these deficiencies. So, to summarize in a formula, AI Agent = LLM (Large Language Model) + Planning + Memory + Tools.

A prompt-based large model is more like a static person—only coming to life when we input into it. On the other hand, the goal of an AI Agent is to be more like a real person. Currently, AI Agents in the field are mainly fine-tuned models based on Meta's open-source Llama 70b or 405b version (with different parameters), possessing the ability to remember and use API access tools. In other aspects, they may require human assistance or input (including interaction and collaboration with other AI Agents). Therefore, we can see that AI Agents in the field mainly exist in the form of KOLs on social networks. To make AI Agents more human-like, integration of planning and action capabilities is necessary, with the sub-item of thought chains in planning being particularly crucial.

III. Chain of Thought (CoT)

The concept of Chain of Thought (CoT) first appeared in Google's 2022 paper titled "Chain-of-Thought Prompting Elicits Reasoning in Large Language Models." The paper suggested that enhancing a model's reasoning ability could be achieved by generating a series of intermediate reasoning steps to help the model better understand and solve complex problems.

A typical CoT Prompt consists of three parts: - A task description with explicit instructions - Logical reasoning supporting the task - Theoretical basis or principles for solving the task - Concrete solution examples This structured approach helps the model understand task requirements, gradually approach the answer through logical reasoning, and thereby improve problem-solving efficiency and accuracy. CoT is particularly suitable for tasks requiring in-depth analysis and multi-step reasoning. While for simple tasks such as mathematical problem-solving or project report writing, CoT may not bring a significant advantage, for complex tasks, it can significantly enhance the model's performance by reducing error rates through a step-by-step problem-solving strategy and improving the quality of task completion.

When building an AI Agent, the CoT plays a key role. The AI Agent needs to understand the received information and make reasonable decisions based on it. CoT helps the Agent effectively process and analyze input information by providing an organized way of thinking, transforming the parsing results into specific action guidelines. This method not only enhances the reliability and efficiency of Agent decisions but also improves the transparency of the decision-making process, making the Agent's behavior more predictable and traceable. By breaking down tasks into multiple small steps, CoT helps the Agent carefully consider each decision point, reducing errors caused by information overload. CoT makes the Agent's decision-making process more transparent, making it easier for users to understand the basis of the Agent's decisions. In interacting with the environment, CoT allows the Agent to continuously learn new information and adjust its behavioral strategy.

As an effective strategy, CoT has not only enhanced the reasoning ability of large language models but has also played a significant role in building more intelligent and reliable AI Agents. By leveraging CoT, researchers and developers can create more adaptive, highly autonomous intelligent systems. CoT has demonstrated its unique advantages in practical applications, especially in handling complex tasks. By breaking tasks down into a series of small steps, it not only improves task accuracy but also enhances the model's interpretability and controllability. This step-by-step problem-solving approach can significantly reduce errors in decision-making when faced with complex tasks due to information overload or complexity. At the same time, this approach improves the traceability and verifiability of the entire solution.

The core function of CoT lies in integrating planning, action, and observation to bridge the gap between reasoning and action. This thinking mode allows AI Agents to effectively devise countermeasures when predicting potential anomalies and accumulate new information while interacting with the external environment to validate pre-set predictions and provide new reasoning bases. CoT acts as a powerful engine of precision and stability, helping AI Agents maintain high operational efficiency in complex environments.

IV. Correct Pseudo-Needs

How exactly should Crypto be integrated with aspects of the AI tech stack? In last year's article, I believed that decentralization of computing power and data was a key step in helping small businesses and individual developers save costs. In this year's Crypto x AI detailed categories by Coinbase, we see a more specific breakdown:

(1) Compute Layer (referring to networks focusing on providing Graphics Processing Unit (GPU) resources for AI developers);

(2) Data Layer (referring to networks supporting decentralized access, orchestration, and validation of the AI data pipeline).

（3）Middleware Layer (referring to the platform or network that supports the development, deployment, and hosting of AI models or agents);

（4）Application Layer (referring to user-facing products that leverage on-chain AI mechanisms, whether B2B or B2C).

Within these four layers of segmentation, each layer has a grand vision, with their goals summarized to combat the Silicon Valley giants' dominance of the next era of the Internet. As I mentioned last year, do we really want to accept the exclusive control of compute power and data by Silicon Valley giants? Under their monopoly, the closed-source large models are black boxes internally. Science, as the most trusted religion of today's humanity, will have every answer from future large models seen as truth by a significant portion of people. But how can this truth be verified? According to the Silicon Valley giants' vision, the permissions ultimately granted to intelligent agents will be unimaginable, such as having the authority to access your wallet, the right to use terminals; how do we ensure people act with good intentions?

Decentralization is the only answer, but sometimes do we need to rationally consider how many will foot the bill for these grand visions? In the past, we could overlook commercial closed-loop situations and compensate for the idealistic errors through Tokens. However, the current situation is very critical. Crypto x AI needs to design with the current reality in mind; for instance, how to balance the supply from both ends in the compute power layer in the case of performance loss and instability? To achieve competitiveness matching centralized clouds. Similarly, how many real users will be in the data layer? How to verify the authenticity and effectiveness of the data provided? What kind of customers need this data? The remaining two layers follow the same principles. In this era, we do not need so many seemingly correct pseudo-demands.

Five, Memes Have Spawned SocialFi

As I mentioned in the first paragraph, Memes have rapidly emerged in a manner that aligns with the Web3 form of SocialFi. Friend.tech was the first Dapp to launch this round of social applications but unfortunately failed due to a rushed token design. Pump.fun has validated the feasibility of pure platformization, without any tokens, without any rules. The demanders and suppliers of attention are unified on the platform; you can post memes, livestream, create coins, comment, trade, everything is free, and Pump.fun only charges a service fee. This is basically the same as the attention economy model found on today's social media platforms like YouTube and Instagram, except with different fee targets; in terms of gameplay, Pump.fun is more Web3.

Base's Clanker is a culmination, benefiting from an integrated ecosystem personally handled by Base, which has its social Dapp as an adjunct, forming a complete internal closed-loop. The intelligent entity Meme is the 2.0 form of a Meme Coin. People always seek novelty, and Pump.fun happens to be at the forefront now. Trends indicate that the whimsical thoughts of silicon-based life forms will replace the crass memes of carbon-based life forms, it's only a matter of time.

I have mentioned Base countless times, but each time the context is different. Looking at the timeline, Base has never been a pioneer, but it has always been a winner.

Six, What Else Can an Intelligent Agent Be?

From a practical perspective, it is impossible for intelligent agents to decentralize for a long time in the future. From the perspective of traditional AI in building intelligent agents, it is not a simple reasoning process that can be solved by decentralization and open source. It needs to access various APIs to reach Web2 content, its operational costs are very high, and the design of a thought chain and the collaboration of multiple intelligent agents usually still rely on a human as an intermediary. We will go through a long transition period until a suitable fusion form emerges, perhaps like UNI. However, as in the previous article, I still believe that intelligent agents will have a significant impact on our industry, just as Cex's presence in our industry, though incorrect, is crucial.

An article called "AI Agent Overview" issued by Stanford & Microsoft last month extensively describes the application of intelligent agents in the medical industry, intelligent machines, and the virtual world. In the appendix of this article, there are already many GPT-4V intelligent agents participating in experimental cases in the development of top 3A games.

There is no need to force the speed of its integration with decentralization too much. I would rather the first puzzle piece that intelligent agents fill in is the ability and speed from the bottom up. We have so many narrative ruins and blank metaverses that need to be filled in by it. At the appropriate stage, we can consider how to make it the next UNI.

References:

What is the ability of the thought chain that emerges from large models? Author: BrainMax

Understanding Agents in One Article, the Next Stop for Large Models. Author: LinguaMind

This article is a contribution and does not represent the views of BlockBeats.