1. The Fickle Nature of Attention

Over the past year, due to a disconnect in application layer narratives, unable to match the pace of infrastructure growth, the crypto space has gradually turned into a competition for attention resources. From Silly Dragon to Goat, from Pump.fun to Clanker, the fickleness of attention has led to a cycle of constant reinvention in this battle. What began with the most conventional form of eye-catching monetization quickly evolved into a platform model that unified attention seekers and providers, ultimately leading to silicon-based life forms becoming new content providers. Among the bizarre array of meme coins, a new entity has emerged that allows retail investors and VCs to reach a consensus: AI Agents.

Attention is ultimately a zero-sum game, though speculation can indeed drive rapid growth. In our previous article on UNI, we revisited the beginning of the last golden age of blockchain, when the explosive growth of DeFi was sparked by Compound Finance’s launch of LP mining. During that era, participating in hundreds, sometimes thousands, of mining pools with yields in the thousands or even tens of thousands of percent APY was the most primitive form of on-chain speculation. Although the outcome was a chaotic collapse of many pools, the influx of “gold rush” miners left unprecedented liquidity in the blockchain space. DeFi eventually broke free from pure speculation and matured into a solid vertical that addressed users’ financial needs in areas such as payments, trading, arbitrage, and staking. AI Agents are currently undergoing a similar “wild growth” phase. What we are exploring now is how crypto can better integrate AI and ultimately elevate the application layer to new heights.

2. How Do AI Agents Operate Autonomously?

In our previous article, we briefly introduced the origins of AI memes through Truth Terminal and explored the future potential of AI Agents. This article will focus on AI Agents themselves.

Let’s start with the definition of an AI Agent. In the AI field, the term “Agent” is an older yet still vague concept, mainly emphasizing autonomy. In other words, any AI that can perceive its environment and make reflexive decisions is considered an Agent. Today, the definition of an AI Agent is closer to that of an intelligent entity, a system designed to mimic human decision-making processes. This system is regarded in academia as the most promising approach toward achieving AGI (Artificial General Intelligence).

In the early versions of GPT, we could clearly sense that the large models were human-like, but when answering complex questions, they often provided answers that were vague or imprecise. The fundamental reason for this was that these models were based on probabilities rather than causality, and they lacked human-like abilities such as tool usage, memory, and planning. AI Agents aim to address these gaps. So, to summarize in a formula: AI Agent = LLM (Large Language Model) + Planning + Memory + Tools.

Prompt-based models are more like a static version of a person, coming to life only when we input data. In contrast, the goal of an AI Agent is to be a more dynamic, human-like entity. Currently, most of the AI Agents in the field are fine-tuned models based on Meta’s open-source Llama 70b or 405b versions (with different parameters), equipped with memory and the ability to use APIs for tool integration. In other areas, they might still need human input or assistance, such as interacting or collaborating with other AI Agents. This is why most AI Agents today primarily exist in the form of KOLs on social networks. To make an AI Agent more human-like, it needs to incorporate planning and action capabilities, with the chain of thought within the planning process being especially crucial.

3. 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 pointed out that by generating a series of intermediate reasoning steps, the model’s reasoning ability could be enhanced, helping it better understand and solve complex problems.

A typical CoT prompt consists of three parts: a task description with clear instructions, a logical basis for the task with the theoretical foundation or principles supporting the solution, and a specific example of the solution. This structured approach helps the model understand the task requirements and, through logical reasoning, gradually approach the answer, improving both the efficiency and accuracy of problem-solving. CoT is particularly suitable for tasks that require deep analysis and multi-step reasoning, such as mathematical problem-solving or writing project reports. For simpler tasks, CoT may not show obvious advantages, but for more complex ones, it can significantly improve the model’s performance by reducing error rates through a step-by-step problem-solving strategy, thus enhancing the quality of task completion.

In the construction of AI Agents, CoT plays a crucial role. AI Agents need to understand the information they receive and make reasonable decisions based on it. CoT provides an ordered thinking process that helps the Agent effectively process and analyze input data, turning the analysis into actionable guidelines. This method not only strengthens the reliability and efficiency of the Agent’s decision-making but also improves the transparency of the decision process, making the Agent’s behavior more predictable and traceable. By breaking tasks down into smaller steps, CoT helps the Agent consider each decision point in detail, reducing errors caused by information overload and making the decision-making process more transparent. This transparency allows users to better understand the basis of the Agent’s decisions. In interactions with the environment, CoT allows the Agent to continuously learn new information and adjust its behavior strategy.

As an effective strategy, CoT not only enhances the reasoning ability of large language models but also plays an important role in building smarter and more reliable AI Agents. By leveraging CoT, researchers and developers can create intelligent systems that are more adaptable to complex environments and highly autonomous. In practical applications, CoT has shown its unique advantages, especially in handling complex tasks. By breaking tasks into a series of smaller steps, it not only improves the accuracy of task resolution but also enhances the model’s interpretability and controllability. This step-by-step problem-solving approach can greatly reduce errors caused by excessive or overly complex information when facing complex tasks. At the same time, this method also improves the traceability and verifiability of the entire solution.

The core function of CoT lies in integrating planning, action, and observation, bridging the gap between reasoning and action. This thinking model allows the AI Agent to devise effective countermeasures when predicting potential anomalies and accumulate new information while interacting with the external environment, validating pre-set predictions and providing new reasoning grounds. CoT acts like a powerful engine of precision and stability, helping the AI Agent maintain high efficiency in complex environments.

4. The Right Kind of Pseudo-Demand

How exactly should Crypto integrate with AI technology stacks? In last year’s article, I suggested that decentralizing computing power and data is a key step in helping small businesses and individual developers save costs. This year, in the detailed breakdown of the Crypto x AI sectors compiled by Coinbase, we can see more specific divisions:

(1) Computing Layer (focused on providing GPU resources for AI developers);

(2) Data Layer (focused on decentralized access, orchestration, and verification of AI data pipelines);

(3) Middleware Layer (platforms or networks supporting the development, deployment, and hosting of AI models or agents);

(4) Application Layer (user-facing products utilizing on-chain AI mechanisms, whether B2B or B2C).

Each of these four layers has grand visions, all of which aim to challenge the domination of Silicon Valley giants in the next era of the internet. As I said last year, do we really need to accept that Silicon Valley giants exclusively control computing power and data? Under their monopolies, closed-source large models are black boxes, and science, as the most revered belief system of humanity today, will rely on the answers given by these large models. But how can these truths be verified? According to the vision of these Silicon Valley giants, the powers held by intelligent agents could exceed our imagination — such as having the authority to make payments from your wallet or control your terminal access. How can we ensure that no malintent arises?

Decentralization is the only answer, but sometimes we need to reasonably consider how many buyers are there for these grand visions. In the past, we could overlook the need for a commercial loop and use Tokens to fill in the gaps caused by idealism. However, the current situation is much more challenging. Crypto x AI must design based on practical circumstances. For example, how do we balance the supply at both ends of the computing layer in cases of performance loss and instability, and still compete with centralized cloud providers? How many real users will the data layer projects actually have? How can we verify the authenticity and validity of the data provided? What types of clients actually need this data? The same logic applies to the other layers. In this era, we don’t need so many seemingly correct pseudo-demands.

5. Meme Has Evolved into SocialFi

As I mentioned in the first section, Meme has rapidly evolved into a Web3-compatible form of SocialFi. Friend.tech was the DApp that fired the first shot in this round of social applications, but unfortunately, it failed due to its rushed Token design. On the other hand, Pump.fun has demonstrated the feasibility of a pure platform model, without any Tokens or rules. The needs of attention seekers and providers converge on this platform, where you can post memes, stream live, mint tokens, comment, trade, and everything is free. Pump.fun only collects a service fee. This model is essentially identical to the attention economy of current social media platforms like YouTube and Instagram, but with a different revenue model and more Web3-centric gameplay.

Base’s Clanker, on the other hand, is the ultimate success story, benefiting from the integrated ecosystem designed by the platform itself. Base has its own social DApp as an auxiliary tool, creating a complete internal loop. The Meme Agent is the 2.0 form of Meme Coin. People are always chasing novelty, and right now, Pump.fun is at the center of attention. From a trend perspective, it’s only a matter of time before silicon-based lifeforms’ whimsical ideas replace the cruder memes of carbon-based lifeforms.

I’ve mentioned Base countless times, with different aspects each time, but one thing remains clear: Base has never been the first mover, but it has always been the winner.

6. What Else Can an AI Agent Be?

From a practical standpoint, AI agents are unlikely to be decentralized in the foreseeable future. In the traditional AI field, building an AI agent is not a problem that can be solved simply through decentralization or open-source processes. AI agents need to connect to various APIs to access Web2 content, and their operating costs are high. The design of the Chain of Thought (CoT) and multi-agent collaboration often still requires human mediation. We will go through a long transition period until we find a suitable form of integration — perhaps something like UNI, but for now, I still believe AI agents will have a significant impact on our industry, much like how CEX exist in our sector — incorrect, but extremely important.

Last month, Stanford & Microsoft published an AI Agent Review that described the applications of AI agents in industries like healthcare, smart machines, and virtual worlds. In the appendix of this paper, there are already numerous experimental cases where GPT-4V, as an AI agent, is participating in the development of top-tier AAA games.

We shouldn’t rush to integrate AI agents with decentralization. What I hope for is that the first puzzle piece AI agents complete is their bottom-up capabilities and speed. There are so many narrative ruins and empty metaverses that need filling, and when the time is right, we can consider how to turn AI agents into the next UNI.

About YBB

YBB is a web3 fund dedicating itself to identify Web3-defining projects with a vision to create a better online habitat for all internet residents. Founded by a group of blockchain believers who have been actively participated in this industry since 2013, YBB is always willing to help early-stage projects to evolve from 0 to 1.We value innovation, self-driven passion, and user-oriented products while recognizing the potential of cryptos and blockchain applications.

Disclaimer:

1. This article is reproduced from [Medium]. The copyright belongs to the original author [YBB]. If there are any objections to the reproduction, please contact the Gate Learn team, who will handle the issue promptly according to the relevant procedures.
2. Disclaimer: The views and opinions expressed in this article are solely those of the author and do not constitute any investment advice.
3. The Gate Learn team translated the article into other languages. Copying, distributing, or plagiarizing the translated articles is prohibited unless mentioned.