PulseAugur
中
实时 07:13:52
English(EN) A chat that did not drift on 400 turns of ambiguous text still left the wrong list for 140 turns

Gemini 3.8 Flash在长对话中存在状态漂移问题;AXI架构提升准确性

一项对比Gemini 3.8 Flash有无AXI架构的测试显示,在400轮对话中,状态管理存在显著差异。单独的Gemini 3.8 Flash实现了99.5%的回复准确率,但一次状态管理失误导致其保持的列表在140轮内不正确。相比之下,包含主机端检查的AXI架构在全部400轮对话中保持了完美的准确性,尽管每个步骤的token数和处理时间略有增加。 AI

影响 突出了LLM在长而模糊的对话中状态持久性方面潜在的问题,并通过主机端架构展示了一个潜在的解决方案。

排序理由 研究论文,详细介绍了对LLM状态管理能力的特定测试。[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 →

Gemini 3.8 Flash在长对话中存在状态漂移问题;AXI架构提升准确性

本文如何被排名

Signal score
44 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
研究论文,详细介绍了对LLM状态管理能力的特定测试。[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
model release, 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]

  1. dev.to — LLM tag TIER_1 English(EN) · Shalendra gaur ·

    一次长达400轮的模糊文本对话,在140轮后仍未偏离,但最终给出了错误列表

    <p><em>This article was created with the help of AI. The AXI architecture is by Rajnish Choubey. The tests were built and run by Shalendra Gaur.</em></p> <p>A fair objection to my graph tests: a graph is rigid and easy to check, which is the easy case for a layer that holds the s…