鏈抽象對於利用鏈上資料至關重要：Chainbase 聯合創始人 Chris Feng

區塊鏈產業受到碎片化和缺乏互通性的困擾。許多領導者和創辦人都強調，這些是大規模採用和創新的最大障礙之一。克里斯·馮（Chris Feng）就是其中之一。

Chris Feng, the Co-founder of Chainbase, an omnichain data network provider, says that the value of on-chain data is largely untapped. He estimates that the number of developers exploiting on-chain data is lesser than the number of developers working at major internet companies.

全鏈數據網路供應商Chainbase的共同創辦人Chris Feng表示，鏈上數據的價值很大程度上尚未開發。他估計，利用鏈上資料的開發人員數量少於在主要網路公司工作的開發人員數量。

Chainbase provides a data interoperability layer for blockchains. It is positioned in the market as an ‘infrastructure provider’. Interoperability and fragmentation are the hottest topics in blockchain right now. Does Chainbase have a solution for these issues?

Chainbase 為區塊鏈提供了資料互通層。它在市場上的定位是「基礎設施提供者」。互通性和碎片化是目前區塊鏈最熱門的話題。 Chainbase 有針對這些問題的解決方案嗎？

In an insightful conversation with The Coin Republic’s Editor-in-Chief, Varuni Trivedi, Feng highlighted the need for chain abstraction. The former consultant discussed the nuances of data decentralization and explained why AI-crypto integration will be normal in the future.

在與《Coin Republic》主編 Varuni Trivedi 的一次富有洞察力的對話中，Feng 強調了鏈抽象的必要性。這位前顧問討論了資料去中心化的細微差別，並解釋了為什麼人工智慧與加密貨幣的整合將在未來成為常態。

Chain Abstraction is Vital for Exploiting On-chain Data

鏈抽象對於利用鏈上資料至關重要

The blockchain industry is plagued with fragmentation and a lack of interoperability. Numerous leaders and founders have emphasized that these are among the biggest barriers to mass adoption and innovation. Chris Feng is one of those voices.

區塊鏈產業受到碎片化和缺乏互通性的困擾。許多領導者和創辦人都強調，這些是大規模採用和創新的最大障礙之一。克里斯·馮（Chris Feng）就是其中之一。

“The blockchain world has created many silos, too many chains and data islands, fragmented information and complex cross-chain operations hinder the ability to fully realize the value of interconnected information. Therefore, Chain Abstraction is a vital topic,” Feng noted.

「區塊鏈世界造成了許多孤島、太多的鏈和資料孤島、碎片化的資訊和複雜的跨鏈操作阻礙了充分實現互聯資訊價值的能力。因此，鏈抽像是一個重要的主題，」馮指出。

Chain abstraction means achieving a high level of interoperability between blockchains. Wallets and decentralized applications (dApps) will work seamlessly (or become accessible) across chains if chain abstraction is achieved. The objective is to make the user’s interaction with the blockchain less daunting. A friendly UX will lead to an increase in adoption.

鏈抽象意味著實現區塊鏈之間的高水準互通性。如果實現了鏈抽象，錢包和去中心化應用程式（dApp）將跨鏈無縫工作（或變得可訪問）。目標是讓用戶與區塊鏈的互動不再那麼令人畏懼。友善的用戶體驗將提高採用率。

As per official documentation, Chainbase is focused on enhancing data interoperability. Data interoperability is a fundamental concept that applies to all blockchains. It is the facility that allows different blockchains to receive and transfer data without special arrangements.

根據官方文檔，Chainbase 致力於增強資料互通性。數據互通性是適用於所有區塊鏈的基本概念。它是允許不同區塊鏈無需特殊安排即可接收和傳輸資料的設施。

“We need to lower the barrier for extracting value from on-chain data. And we also need to bring more diverse data to the chain to create new possibilities.”

「我們需要降低從鏈上資料中提取價值的障礙。而且我們還需要將更多樣化的數據帶入鏈上，創造新的可能性。

Theia: a Crypto World Model

Theia：加密世界模型

On August 15th, 2024, Chainbase launched its ‘milestone’ project, Theia. Theia is an AI-powered ‘Crypto World Model’ that interprets on-chain data. The model’s chatbot answers questions about on-chain data. Theia is fed data from Coinmarketcap (whitepapers, official blog posts, and news articles) of the top 2000 projects by market capitalization and credible research reports on these projects.

2024 年 8 月 15 日，Chainbase 啟動了其「里程碑」專案 Theia。 Theia 是一個由人工智慧驅動的“加密世界模型”，可以解釋鏈上數據。該模型的聊天機器人回答有關鏈上數據的問題。 Theia 從 Coinmarketcap（白皮書、官方部落格文章和新聞文章）取得市值排名前 2000 個項目的數據以及這些項目的可信研究報告。

World Models are alternatives to Large Language Models (LLMs). Feng pointed out that LLMs are useful for human-like interaction but fall short against mathematical questions and transaction queries. This is where World Models shine. World Models are designed to capture knowledge from a world or an environment to simulate that environment. 

世界模型是大型語言模型 (LLM) 的替代方案。馮指出，法學碩士對於類人互動很有用，但在解決數學問題和交易查詢方面卻有所不足。這就是世界模特兒的閃光點。世界模型旨在從世界或環境中捕獲知識來模擬該環境。

“…world models aim to simulate and understand the external world to perform tasks and make decisions in complex settings,” Feng explained. He added that while world models will act like the user’s digital twin, LLMs are limited to being assistants.

「......世界模型旨在模擬和理解外部世界，以在複雜的環境中執行任務並做出決策，」馮解釋道。他補充說，雖然世界模型將充當用戶的數位雙胞胎，但法學碩士僅限於助手。

“The key here is to simulate the environment before executing the task. So in a specialized field like crypto, where there is a wealth of domain-specific knowledge not commonly found in social media and where data is not just plain text or images, LLMs can be significantly hindered. LLMs are often limited to task planning without the capability to extend further or make close loop decisions.”

「這裡的關鍵是在執行任務之前模擬環境。因此，在像加密貨幣這樣的專業領域，那裡有大量在社交媒體中不常見的特定領域知識，而且數據不僅僅是純文字或圖像，法學碩士可能會受到嚴重阻礙。法學碩士通常僅限於任務規劃，無法進一步擴展或做出閉環決策。

Theia is a one-of-a-kind dedicated blockchain Chatbot. It gathers data on transactions, smart contracts, and NFTs, among other on-chain data assets. Theia is equipped with multiple data analysis tools to help identify patterns. The AI model records and learns from user queries and user feedback.

Theia 是一款獨一無二的專用區塊鏈聊天機器人。它收集交易、智能合約和 NFT 以及其他鏈上資料資產的資料。 Theia 配備了多種數據分析工具來幫助識別模式。人工智慧模型記錄使用者查詢和使用者回饋並從中學習。

The official full name of the first version is Theia-Llama-3.1-8B; as per Chainbase, Theia-Llama-3.1-8B is fine-tuned from the base model Llama-3.1-8B-Instruct. It was open-sourced on October 11th and developed especially for the crypto sector. 

第一版的官方全名為Theia-Llama-3.1-8B；根據 Chainbase，Theia-Llama-3.1-8B 是在基礎模型 Llama-3.1-8B-Instruct 的基礎上進行微調的。它於 10 月 11 日開源，專為加密貨幣領域開發。

Notably, Theia-Llama-3.1-8B outperformed leading LLMs such as ChatGPT-4o, Claude-3-Sonet, Gemini-1.5-Pro, and Llama-3.1-8B-Instruct, against Chainbase’s benchmark for Crypto AI models.

值得注意的是，根據Chainbase 的加密AI 模型基準，Theia-Llama-3.1-8B 的表現優於ChatGPT-4o、Claude-3-Sonet、Gemini-1.5-Pro 和Llama-3.1-8B-Instruct 等領先的LLM。

Are LLMs Overrated?

LLM 是否被高估了？

LLMs have been among the most lucrative applications of AI in the past three years. OpenAI’s ChatGPT, Gemini, Llama, and other models have dominated discussions and even the news from the AI beat. Major flaws in these models have been spotted. Chris explains where these models fall short and how World Models are different.

過去三年來，法學碩士一直是人工智慧最賺錢的應用之一。 OpenAI 的 ChatGPT、Gemini、Llama 和其他模型主導了討論，甚至是 AI 領域的新聞。這些模型的主要缺陷已經被發現。 Chris 解釋了這些模型的缺點以及世界模型的差異。

Feng highlights two big challenges that LLMs face. First is the lack of high-quality datasets (on-chain data). “There are a lot of projects in the on-

馮強調了法學碩士面臨的兩大挑戰。首先是缺乏高品質的數據集（鏈上數據）。 「目前有很多項目