Article

HierMem: Context Curation Over Context Scaling — Hierarchical Memory with Invariant Constraint Placement for Long-Horizon LLM Conversations

2026-06-17

Abstract excerpt

<title>Abstract</title> <p>The industry assumes that larger context windows are the solution to long-horizon conversational AI. This assumption is fundamentally flawed. Flooding a Large Language Model (LLM) with raw historical turns actively degrades reasoning through the "lost in the middle" effect, dilutes critical user constraints, and incurs massive compute costs. LLMs do not need to process every past intera...

Topics

Open a Topic to create a Post that cites this publication.

Uncategorized literature

Identifiers and source

Literature Corpus work: f51be69b-e1a6-5c5d-a7ac-3a9c7ef5510f
DOI: 10.21203/rs.3.rs-10055780/v1

Open publication

Related research

Semantic proximity does not establish scientific evidence.

Click a neighbor to travelStep 1 · 12 closest

HierMem: Context Curation Over Context Scaling — Hierarchical Memory with Invariant Constraint Placement for Long-Horizon LLM ConversationsDOI 10.21203/rs.3.rs-10055780/v1Read

Reverse Prompting: A Memory-Efficient Paradigm for LLM Agents Through Structural RegenerationSemantic proximity Read

Dimension-Direct Routing: Achieving 25% Depth Improvement in Multi- Model LLM Systems via Explicit Capability FactorizationSemantic proximity Read

Exploring the Concept of Dynamic Memory Persistence in Large Language Models for Optimized Contextual ComprehensionSemantic proximity Read

Neural Re-Contextualization for Dynamic Semantic Control in Large Language ModelsSemantic proximity Read

iRAT: Replanning and Controlled Retrieval for Robust LLM ReasoningSemantic proximity Read

Reasoning in Large Language Models: From Chain-of-Thought to Massively Decomposed Agentic ProcessesSemantic proximity Read

Dynamic Adaptive Reasoning: Optimizing LLM Inference-Time Thinking via Intent-Aware SchedulingSemantic proximity Read

Domain-Partitioned Retrieval as a Hallucination Mitigation Strategy in Conversational RAG: The PC-RAG ArchitectureSemantic proximity Read

Beyond Words: The Evolution of Large Language Models in Context-AwareSemantic proximity Read

iRAT: Improved Retrieval-Augmented Thinking for Context-Aware Replanning-Based ReasoningSemantic proximity Read

Context-Grounded Factuality Enhancement in LLM Responses via Multi-Stage Critique and RefinementSemantic proximity Read

Basics of Large Language Models - transformers to LLMsSemantic proximity Read