# Improving Complex Reasoning with Dynamic Prompt Corruption: A soft prompt Optimization Approach

arXiv: https://arxiv.org/abs/2503.13208 · HTML: https://mlsanity.com/paper/2503.13208/

## Key facts

- Cited by 9 papers in the map (data through 14 Oct 2026).
- Placed in Prompting and chain-of-thought reasoning, one of 17 directions in LLM reasoning.
- Builds on 13 of its 36 references that are in the map.
- First posted on arXiv 17 Mar 2025.

## Placed in

- LLM reasoning / [Prompting and chain-of-thought reasoning](https://mlsanity.com/lr/prompting-and-chain-of-thought-reasoning/)

## Builds on (most cited first, 13 of 13)

- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Llama 2: Open Foundation and Fine-Tuned Chat Models](https://mlsanity.com/paper/2307.09288/) (2023)
- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [Measuring Mathematical Problem Solving With the MATH Dataset](https://mlsanity.com/paper/2103.03874/) (2021)
- [The Power of Scale for Parameter-Efficient Prompt Tuning](https://mlsanity.com/paper/2104.08691/) (2021)
- [Mistral 7B](https://mlsanity.com/paper/2310.06825/) (2023)
- [Efficient Streaming Language Models with Attention Sinks](https://mlsanity.com/paper/2309.17453/) (2023)
- [GPT Understands, Too](https://mlsanity.com/paper/2103.10385/) (2021)
- [Parameter-Efficient Fine-Tuning for Large Models: A Comprehensive Survey](https://mlsanity.com/paper/2403.14608/) (2024)
- [Unveiling and Harnessing Hidden Attention Sinks: Enhancing Large Language Models without Training through Attention Calibration](https://mlsanity.com/paper/2406.15765/) (2024)
- [Instance-adaptive Zero-shot Chain-of-Thought Prompting](https://mlsanity.com/paper/2409.20441/) (2024)
- [Exploring Universal Intrinsic Task Subspace via Prompt Tuning](https://mlsanity.com/paper/2110.07867/) (2021)
- [Dynamic Prompting: A Unified Framework for Prompt Tuning](https://mlsanity.com/paper/2303.02909/) (2023)

## Built on by (most cited first, 9 of 9)

- [Linking Process to Outcome: Conditional Reward Modeling for LLM Reasoning](https://mlsanity.com/paper/2509.26578/) (2025)
- [Prompt-R1: Collaborative Automatic Prompting Framework via End-to-end Reinforcement Learning](https://mlsanity.com/paper/2511.01016/) (2025)
- [Not All Tokens and Heads Are Equally Important: Dual-Level Attention Intervention for Hallucination Mitigation](https://mlsanity.com/paper/2506.12609/) (2025)
- [SIRI-Bench: Challenging VLMs' Spatial Intelligence through Complex Reasoning Tasks](https://mlsanity.com/paper/2506.14512/) (2025)
- [GeoCAD: Local Geometry-Controllable CAD Generation with Large Language Models](https://mlsanity.com/paper/2506.10337/) (2025)
- [SoftSkill: Behavioral Compression for Contextual Adaptation](https://mlsanity.com/paper/2606.20333/) (2026)
- [Multimodal Policy Internalization for Conversational Agents](https://mlsanity.com/paper/2510.09474/) (2025)
- [Reasoning Fails Where Step Flow Breaks](https://mlsanity.com/paper/2604.06695/) (2026)
- [Optimizing Soft Prompt Tuning via Structural Evolution](https://mlsanity.com/paper/2602.16500/) (2026)

## Abstract (arXiv)

Prompt-tuning (PT) for large language models (LLMs) can facilitate the performance on various conventional NLP tasks with significantly fewer trainable parameters. However, our investigation reveals that PT provides limited improvement and may even degrade the primitive performance of LLMs on complex reasoning tasks. Such a phenomenon suggests that soft prompts can positively impact certain instances while negatively affecting others, particularly during the later phases of reasoning. To address these challenges, We first identify an information accumulation within the soft prompts. Through detailed analysis, we demonstrate that this phenomenon is often accompanied by erroneous information flow patterns in the deeper layers of the model, which ultimately lead to incorrect reasoning outcomes. we propose a novel method called Dynamic Prompt Corruption (DPC) to take better advantage of soft prompts in complex reasoning tasks, which dynamically adjusts the influence of soft prompts based on their impact on the reasoning process. Specifically, DPC consists of two stages: Dynamic Trigger and Dynamic Corruption. First, Dynamic Trigger measures the impact of soft prompts, identifying whether beneficial or detrimental. Then, Dynamic Corruption mitigates the negative effects of soft prompts by selectively masking key tokens that interfere with the reasoning process. We validate the proposed approach through extensive experiments on various LLMs and reasoning tasks, including GSM8K, MATH, and AQuA. Experimental results demonstrate that DPC can consistently enhance the performance of PT, achieving 4%-8% accuracy gains compared to vanilla prompt tuning, highlighting the effectiveness of our approach and its potential to enhance complex reasoning in LLMs.