# Do LLMs Benefit From Their Own Words?

arXiv: https://arxiv.org/abs/2602.24287 · HTML: https://mlsanity.com/paper/2602.24287/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- Builds on 25 of its 31 references that are in the map.
- First posted on arXiv 27 Feb 2026.

## Builds on (most cited first, 25 of 25)

- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness](https://mlsanity.com/paper/2205.14135/) (2022)
- [Lost in the Middle: How Language Models Use Long Contexts](https://mlsanity.com/paper/2307.03172/) (2023)
- [gpt-oss-120b & gpt-oss-20b Model Card](https://mlsanity.com/paper/2508.10925/) (2025)
- [DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models](https://mlsanity.com/paper/2512.02556/) (2025)
- [WildChat: 1M ChatGPT Interaction Logs in the Wild](https://mlsanity.com/paper/2405.01470/) (2024)
- [LLMs Get Lost In Multi-Turn Conversation](https://mlsanity.com/paper/2505.06120/) (2025)
- [LLMLingua-2: Data Distillation for Efficient and Faithful Task-Agnostic Prompt Compression](https://mlsanity.com/paper/2403.12968/) (2024)
- [LLMLingua: Compressing Prompts for Accelerated Inference of Large Language Models](https://mlsanity.com/paper/2310.05736/) (2023)
- [Trae Agent: An LLM-based Agent for Software Engineering with Test-time Scaling](https://mlsanity.com/paper/2507.23370/) (2025)
- [Cartridges: Lightweight and general-purpose long context representations via self-study](https://mlsanity.com/paper/2506.06266/) (2025)
- [Learning to Filter Context for Retrieval-Augmented Generation](https://mlsanity.com/paper/2311.08377/) (2023)
- [Unlocking Context Constraints of LLMs: Enhancing Context Efficiency of LLMs with Self-Information-Based Content Filtering](https://mlsanity.com/paper/2304.12102/) (2023)
- [Learning to Compress Prompt in Natural Language Formats](https://mlsanity.com/paper/2402.18700/) (2024)
- [Reducing Cost of LLM Agents with Trajectory Reduction](https://mlsanity.com/paper/2509.23586/) (2025)
- [The Complexity Trap: Simple Observation Masking Is as Efficient as LLM Summarization for Agent Context Management](https://mlsanity.com/paper/2508.21433/) (2025)
- [PENCIL: Long Thoughts with Short Memory](https://mlsanity.com/paper/2503.14337/) (2025)
- [Attentive History Selection for Conversational Question Answering](https://mlsanity.com/paper/1908.09456/) (2019)
- [Provence: efficient and robust context pruning for retrieval-augmented generation](https://mlsanity.com/paper/2501.16214/) (2025)
- [Lost in the Noise: How Reasoning Models Fail with Contextual Distractors](https://mlsanity.com/paper/2601.07226/) (2026)
- [CoHS-CQG: Context and History Selection for Conversational Question Generation](https://mlsanity.com/paper/2209.06652/) (2022)
- [The ShareLM Collection and Plugin: Contributing Human-Model Chats for the Benefit of the Community](https://mlsanity.com/paper/2408.08291/) (2024)
- [ERGO: Entropy-guided Resetting for Generation Optimization in Multi-turn Language Models](https://mlsanity.com/paper/2510.14077/) (2025)
- [Learning to Select the Relevant History Turns in Conversational Question Answering](https://mlsanity.com/paper/2308.02294/) (2023)
- [Haystack Engineering: Context Engineering for Heterogeneous and Agentic Long-Context Evaluation](https://mlsanity.com/paper/2510.07414/) (2025)

## Built on by (most cited first, 5 of 5)

- [Verbalizing LLMs' assumptions to explain and control sycophancy](https://mlsanity.com/paper/2604.03058/) (2026)
- [Markovian Generation Chains in Large Language Models](https://mlsanity.com/paper/2603.11228/) (2026)
- [Pingquanqi (Equalizer): A Cross-Domain Sociotechnical Framework for Human-Agent Interaction Governance](https://mlsanity.com/paper/2606.26573/) (2026)
- [Rethinking Multi-Agent Collaboration: When More Is Less](https://mlsanity.com/paper/2609.19759/) (2026)
- [Where Did We Go Wrong? Localizing Errors to Repair Multi-Turn Conversations](https://mlsanity.com/paper/2610.15154/) (2026)

## Abstract (arXiv)

In multi-turn conversations, large language models typically condition on the full conversation history: both past user prompts and assistant responses. We revisit this design choice by comparing full-context prompting to four alternative, substantially-reduced context configurations. Analyzing in-the-wild multi-turn conversations across three open reasoning and one state-of-the-art model, we find that response quality is largely preserved under aggressive context filtering: replacing all prior assistant turns with one-sentence summaries or keeping only the most recent user--assistant exchange often matches storing full context in performance while using roughly 8x less context. To understand this result, we observe that a substantial fraction of user turns (36.4%) in multi-turn conversations are self-contained and that many follow-up turns can be addressed by seeing only the immediately preceding user--assistant exchange. Furthermore, we find that when models condition on their own past responses, this can lead to context pollution, a phenomenon in which reasoning errors, hallucinations, or stylistic artifacts propagate across turns. Motivated by these findings, we design a context-filtering approach that selectively omits the assistant-side history. Taken together, these findings suggest moving away from storing full dialogue transcripts and instead retaining only what is relevant.