Papers with Interpretability analysis
Transferable Direct Prompt Injection via Activation-Guided MCMC Sampling (2025.emnlp-main)
Copied to clipboard
| Challenge: | Experimental results show superior cross-model transferability . Prompt injection attacks are among the most critical threats . |
| Approach: | They propose an activations-guided prompt injection attack framework to address the impracticality of existing white-box/gray-box methods and the poor transferability of black-box approaches. |
| Outcome: | The proposed framework achieves 49.6% success rate and 34.6% improvement over human-crafted prompts on five mainstream LLMs. |
Non-Existent Relationship: Fact-Aware Multi-Level Machine-Generated Text Detection (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for detecting LLMs lack the authenticity of the entity graph . lmgenerated text is misused, including fake news and spam . |
| Approach: | They propose a fact-aware model that assesses discrepancies between textual and factual entity graphs through graph comparison. |
| Outcome: | The proposed model outperforms state-of-the-art methods on three public datasets showing that it can capture differences in entity graphs between machine-generated and human-written texts. |
Learning Temporally-Aware Sample Weights for Preference Optimization (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for preference optimization rely on static functions of instantaneous model states and ignore temporal learning dynamics. |
| Approach: | They propose a framework that meta-learns adaptive weights using three temporal features: reward margin evolution, learning volatility, and reference deviation. |
| Outcome: | The proposed framework achieves statistically significant improvements over baselines on models ranging from 7B to 70B parameters. |