ASI-1 mini：fetch.ai公布了世界上第一个Web3本地大语模型（LLM）

随着AI X Web3市场的继续扩展，分散的AI和自主代理技术提供商Fetch.ai最近宣布了ASI-1 Mini的推出

Announced: Fetch.ai Unveils ASI-1 Mini, Web3's Initial Large Language Model

宣布：fetch.ai揭开了Web3的最初大语言模型ASI-1 mini

In a recent development, Fetch.ai, a provider of autonomous agent technology and decentralized AI, has announced the launch of ASI-1 Mini, a Web3-native large language model (LLM) designed to prioritize autonomous agent workflows. This marks a significant step for the Artificial Superintelligence (ASI) Alliance, a group founded by Fetch.ai alongside prominent Web3 entities like SingularityNET, Ocean Protocol, and CUDOS.

在最近的开发中，自主代理技术和分散AI的提供商Fetch.ai宣布启动ASI-1 Mini，这是一种旨在优先考虑自主座席工作流的Web3-本地大型语言模型（LLM）。这标志着人工超级智能（ASI）联盟的重要一步，这是一个由fetch.ai与著名的Web3实体（如SingularityNet，Ocean Protocent和Cudos）共同创立的群体。

ASI-1 Mini serves as the first model in ASI's broader "ASI:" family, with more advanced models slated for release in the near future under the Cortex group. Notably, ASI-1 Mini is designed to operate efficiently on just two GPUs, thanks to its innovative architecture. This reportedly achieves an eight-fold improvement in hardware efficiency compared to existing solutions, significantly reducing the infrastructure costs associated with deploying enterprise-grade AI systems and making them more accessible to a wider range of organizations and developers.

ASI-1 Mini是ASI更广泛的“ ASI”家庭中的第一个模型，其中更先进的模型计划在不久的将来在Cortex组下发布。值得注意的是，ASI-1 MINI旨在仅在两个GPU上有效地运行，这要归功于其创新的架构。据报道，与现有解决方案相比，这可以大大降低与部署企业级AI系统相关的基础架构成本，并使它们更容易被更广泛的组织和开发人员访问。

A New Era: Web3-Native AI Architecture and Ownership

一个新时代：Web3本地AI架构和所有权

On a technical level, ASI-1 Mini's architecture not only incorporates the traditional Mixture of Experts (MoE) framework but also adds what Fetch.ai refers to as a Mixture of Models (MoM) and Mixture of Agents (MoA) approach, enabling it to dynamically select and utilize specialized components for different tasks. Commenting on the development, Humayun Sheikh, CEO of Fetch.ai and Chairman of the ASI Alliance, stated:

在技​​术层面上，ASI-1 Mini的建筑不仅包含了专家（MOE）框架的传统混合物，而且还添加了Fetch.AI所指的内容（MOM）和代理（MOA）方法的混合物，还可以启用它动态选择并利用专用组件来进行不同的任务。 Fetch.ai首席执行官兼ASI联盟主席Humayun Sheikh表示：

"ASI-1 Mini is just the start, over the coming days, we will be rolling out advanced agentic tool-calling, expanded multi-modal capabilities, and deeper Web3 integrations. With these enhancements, ASI-1 Mini will drive agentic automation while ensuring that AI’s value creation remains in the hands of those who fuel its growth.”

“ ASI-1 Mini仅仅是开始，在接下来的几天里，我们将推出高级代理工具称呼，扩展的多模式功能和更深的Web3集成。通过这些增强功能，ASI-1 Mini将驱动代理自动化，而同时可以驱动代理自动化。确保人工智能的价值创造仍然掌握在那些推动其增长的人的手中。”

Moreover, the model will integrate seamlessly with a multitude of Web3 wallets and operate using $FET tokens, allowing users to not only utilize the AI but also potentially benefit from its growth and development. Through the ASI: platform, community members can participate in model training and development, thus sharing in the financial rewards generated by these systems.

此外，该模型将与众多Web3钱包无缝集成并使用$ FET令牌运营，从而使用户不仅可以利用AI，而且还可以从其增长和开发中受益。通过ASI：平台，社区成员可以参与模型培训和开发，从而分享这些系统产生的财务奖励。

Performance Metrics and More

性能指标和更多

Early performance benchmarks indicate that ASI-1 Mini performs competitively with leading LLMs in specialized domains. Notably, the model boasts four dynamic reasoning modes: Multi-Step, Complete, Optimized, and Short Reasoning, which can be switched between based on task requirements. A key focus of ASI-1 Mini's development has been addressing the "black box" challenge faced by AI systems, where they often provide outputs without clear explanations for their reasoning process.

早期性能基准表明，ASI-1 MINI在专用域中与领先的LLM竞争性能。值得注意的是，该模型具有四种动态推理模式：多步，完整，优化和简短的推理，可以根据任务要求在之间切换。 ASI-1 Mini开发的重点是应对AI系统面临的“黑匣子”挑战，在该挑战中，他们经常在没有明确解释其推理过程的情况下提供输出。

To tackle this issue, ASI-1 Mini employs what the company describes as continuous multi-step reasoning. Unlike conventional models that typically reason only at the beginning of a task, ASI-1 Mini maintains an ongoing reasoning process throughout its operations, enabling real-time adjustments/corrections and greater insights into how the model arrives at its conclusions.

为了解决这个问题，ASI-1 MINI采用了公司所说的连续多步推理。与通常仅在任务开始时的常规模型不同，ASI-1 Mini在整个操作过程中都保持了一个持续的推理过程，从而实现了实时调整/更正，并对模型如何得出其结论有了更多的了解。

This transparency initiative is further supported by the system's three-layered architecture. The foundational layer, powered by ASI-1 Mini, serves as the central intelligence hub, while the specialization layer houses domain-specific models, and the action layer manages execution through specialized agents.

该系统的三层体系结构进一步支持了该透明度计划。由ASI-1 MINI提供支持的基础层是中央智能中心，而专业层则容纳了域特异性模型，而动作层通过专业的代理来管理执行。

As the technology continues to advance and additional features are implemented, the broader impact of Fetch.ai's Web3-native approach to AI development will become clearer in the near term. For now, ASI-1 Mini marks a significant step in this direction, combining advanced AI capabilities with decentralized ownership and development models at scale.

随着技术的不断发展并实施了其他功能，Fetch.AI的Web3-native AI开发方法的广泛影响将在短期内变得更加清晰。目前，ASI-1迷你标志着在这个方向上迈出的重要一步，将先进的AI功能与分散的所有权和开发模型相结合。