使用Jev和LangGraph构建生产级智能体

LangChain Blog ·

对Jev和LangGraph的技术概述,涵盖前沿大语言模型(LLM)的成本与性能权衡、窄决策任务的基准测试结果,以及AI软件架构的三种类型:传统方法、基于智能体的方法、AI驱动方法(代码与模型混合)。 阅读 3 条观点,查看支持证据与原始来源。

Sydney Runkle, Hunter Lovell

理解这篇

3 个要点

综合解读

  1. 被当作‘万能神’的前沿LLM既昂贵又缓慢

    前沿大语言模型(LLM)正日益被当作‘万能神’来使用——任何模糊输入的任务都交给它处理——但这种通用性导致每次调用都产生高昂成本和明显延迟,哪怕只需一个简单的二元决策。

    支持这项说法 1

    随着前沿LLM的能力越来越强,我们开始把它们当作“神”,在任何涉及模糊输入的任务中都去求助于它们:散文生成、文档提取、搜索、排序、研究、分类等等。不幸的是,“神”既昂贵又缓慢。你在每次调用时都要为这种通用性付出代价,即使你需要的只是一个“是”或“否”。

    Sydney Runkle, Hunter Lovell · 段落 2

    原始摘录
    As frontier LLMs have gotten more capable, we've started treating them like god, reaching for them for any task with fuzzy inputs: prose generation, document extraction, search, ranking, research, classification, and more. Unfortunately, god is expensive, and slow. You pay for that generality on every call, even when all you needed was a yes or no.
    回到原文语境 →
  2. Jev在窄决策任务上的基准测试速度最高快200倍、成本低400倍

    在路由、分类等窄决策任务(这类任务在智能体架构中十分常见)上,TypeSafe的基准测试显示:相比当前领先的LLM,Jev的运行速度最高可达200倍,单位成本低至1/400。

    支持这项说法 1

    这就是为什么Jev的发布引起了如此多的关注。在诸如许多智能体所围绕的路由和分类步骤这类窄决策任务上,TypeSafe的基准测试显示,Jev的运行速度比领先的LLM快多达200倍,成本低多达400倍。

    Sydney Runkle, Hunter Lovell · 段落 3

    原始摘录
    That's why Jev's launch drew so much attention. On narrow decision tasks, like the routing and classification steps many agents are built around, TypeSafe's benchmarks show Jev running up to 200x faster and 400x cheaper than leading LLMs.
    回到原文语境 →
  3. AI驱动的软件在代码与模型之间分配控制权

    AI驱动的软件采取了一条中间路线:确定性代码保留结构控制和精确计算,而模型仅在决策分支处处理语义判断——这不同于传统软件(完全显式)或智能体(完全由模型驱动)。

    支持这项说法 1

    TypeSafe的文档列出了构建软件的三种方式。传统软件由显式逻辑构成,每个分支都由人工编写且可审计,但也因此显得僵化。智能体则走向了另一个极端:一个模型在每一步都包揽决策,控制流从代码转移到了提示词中。AI驱动的软件采取了中间路线。代码保留结构并处理精确计算,而模型只出现在需要语义判断的分支处。

    Sydney Runkle, Hunter Lovell · 段落 7

    原始摘录
    TypeSafe's docs lay out three ways to build software. Traditional software is made of explicit logic, with every branch written out by hand and auditable, but rigid as a result. Agents swung the other way: one model juggles decisions at every step, and control flow moves out of code and into prompts. AI-powered software takes the middle path. Code keeps the structure and handles exact computation, and a model sits only at the branches that need semantic judgment.
    回到原文语境 →

    继续探索

    AI软件架构 →

关键段落3

带明确归属与语境的原文片段。打开原始文本核查出处。

AI模型成本与性能

被当作‘万能神’的前沿LLM既昂贵又缓慢

随着前沿LLM的能力越来越强,我们开始把它们当作“神”,在任何涉及模糊输入的任务中都去求助于它们:散文生成、文档提取、搜索、排序、研究、分类等等。不幸的是,“神”既昂贵又缓慢。你在每次调用时都要为这种通用性付出代价,即使你需要的只是一个“是”或“否”。

原始摘录
As frontier LLMs have gotten more capable, we've started treating them like god, reaching for them for any task with fuzzy inputs: prose generation, document extraction, search, ranking, research, classification, and more. Unfortunately, god is expensive, and slow. You pay for that generality on every call, even when all you needed was a yes or no.
AI模型基准测试

Jev在窄决策任务上的基准测试速度最高快200倍、成本低400倍

这就是为什么Jev的发布引起了如此多的关注。在诸如许多智能体所围绕的路由和分类步骤这类窄决策任务上,TypeSafe的基准测试显示,Jev的运行速度比领先的LLM快多达200倍,成本低多达400倍。

原始摘录
That's why Jev's launch drew so much attention. On narrow decision tasks, like the routing and classification steps many agents are built around, TypeSafe's benchmarks show Jev running up to 200x faster and 400x cheaper than leading LLMs.
AI软件架构

AI驱动的软件在代码与模型之间分配控制权

TypeSafe的文档列出了构建软件的三种方式。传统软件由显式逻辑构成,每个分支都由人工编写且可审计,但也因此显得僵化。智能体则走向了另一个极端:一个模型在每一步都包揽决策,控制流从代码转移到了提示词中。AI驱动的软件采取了中间路线。代码保留结构并处理精确计算,而模型只出现在需要语义判断的分支处。

原始摘录
TypeSafe's docs lay out three ways to build software. Traditional software is made of explicit logic, with every branch written out by hand and auditable, but rigid as a result. Agents swung the other way: one model juggles decisions at every step, and control flow moves out of code and into prompts. AI-powered software takes the middle path. Code keeps the structure and handles exact computation, and a model sits only at the branches that need semantic judgment.

来源与研究方法

这些观点均关联原始来源。转述已明确标注,不作为逐字原话展示。

打开转录或来源材料 (在新标签页中打开)报告问题