---
title: "ImplicitHop: Evaluating knowledge integration across context and memory"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2610.41169v1"
permalink: "https://feed.astralor.com/items/itm_6a39a158f2a21558675f57ab"
published: "2026-10-30T02:44:51.000Z"
score: 56
channel: "tech"
tags: ["LLM","AI Agent","推理效率","ImplicitHop","Evaluating","LLMs","Whether"]
---

# ImplicitHop: Evaluating knowledge integration across context and memory

## 速览

The ability to integrate new information and deduce its consequences is central to learning, understanding, and reasoning in large language models (LLMs). Whether information arrives during training or in-context, model…

## 关键点

- The ability to integrate new information and deduce its consequences is central to learning, understanding, and reasoning in large language models (LLMs).
- Whether information arrives during training or in-context, models must not only store local facts, but propagate their long-range implications globally.
- To explore information integration in LLMs, we introduce two complementary natural language benchmarks for large-scale evaluation of deductive reasoning as a function of inference depth.

## 全文

The ability to integrate new information and deduce its consequences is central to learning, understanding, and reasoning in large language models (LLMs). Whether information arrives during training or in-context, models must not only store local facts, but propagate their long-range implications globally. To explore information integration in LLMs, we introduce two complementary natural language benchmarks for large-scale evaluation of deductive reasoning as a function of inference depth. The first is a formal graph benchmark enabling controlled evaluation over a wide range of graph topologies and integration tasks; the second is a knowledge composition benchmark grounded in real-world knowledge graphs. We use these benchmarks to compare knowledge integration from information encoded in context versus information encoded in model weights during pretraining, evaluating a set of open models implicitly, without chain of thought. We find that accuracy decays sharply with increased composition length for both pretrained and in-context knowledge. Information presented in-context, however, substantially outperforms deduction over information encoded in weights. These results quantify a fundamental limitation of current models with direct implications for prompt-based control of agent behavior and for continual learning strategies, and they provide a foundation for developing training, prompting, and integration methods that support deeper reasoning.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_6a39a158f2a21558675f57ab