English(EN) Your LLM agent re-reads its history every step. We measured what that costs

新的AXI架构大幅降低了LLM代理的成本和错误

作者 PulseAugur 编辑部 · [1 个来源] · 2026-11-11 15:16

由Rajnish Choubey开发并由Shalendra Gaur测试的新架构AXI，显著提高了LLM代理的性能和可靠性。AXI解决了LLM代理在每一步都重新读取其全部历史记录的问题，这会导致长任务的成本和错误呈指数级增长。通过将确切状态存储在主机上，并为LLM提供简洁、特定于步骤的提示，AXI将提示大小减少了160多倍，从而在延迟、吞吐量和成本效益方面取得了显著的改进。这种方法还大大降低了错误率，确保了在传统基于聊天的LLM代理失败的任务上的完成准确性。 AI

影响该架构可以显著降低长任务LLM代理的运营成本并提高其可靠性。

排序理由研究论文，详细介绍了一种新的LLM代理架构。[lever_c_demoted from research: ic=1 ai=1.0]

在 dev.to — LLM tag 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 34 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 研究论文，详细介绍了一种新的LLM代理架构。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: infra, product; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: Breaking (< 6h); Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

dev.to — LLM tag TIER_1 English(EN) · Shalendra gaur · 2026-11-11 15:16

你的 LLM 代理每一步都会重读其历史记录。我们测量了这会付出什么代价

<p>At turn 223, a Gemini 3.8 Flash chat started making mistakes. By turn 287 it was wrong. The same model, on the same job, with the same 1,500 steps, made none.</p> <p>The difference was not the model. It was what the model was handed at each step.</p> <h2> The cost nobody price…

本文如何被排名

报道来源 [1]

你的 LLM 代理每一步都会重读其历史记录。我们测量了这会付出什么代价

相关实体

相关话题