Papers by Kimin Lee
Improbable Bigrams Expose Vulnerabilities of Incomplete Tokens in Byte-Level Tokenizers (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent studies have revealed that tokenizers can be exploited to elicit unwanted behavior. |
| Approach: | They propose to exploit incomplete tokens with stray bytes to exploit their dependency . they propose to use improbable bigrams to exploit the dependency of their adjacent tokens . |
| Outcome: | The proposed tokenizers can be exploited to elicit unwanted behavior in language models. |
Understanding Impact of Human Feedback via Influence Functions (2025.acl-long)
Copied to clipboard
| Challenge: | In reinforcement learning from human feedback, human feedback can be noisy, inconsistent or biased . this variability can lead to misaligned reward signals, potentially causing unintended side effects . |
| Approach: | They propose an approximation method that measures the impact of human feedback on the performance of reward models. |
| Outcome: | The proposed method detects common labeler biases in human feedback datasets and guides labelers in refining their strategies to better align with expert feedback. |
What Really Matters in Many-Shot Attacks? An Empirical Study of Long-Context Vulnerabilities in LLMs (2025.acl-long)
Copied to clipboard
| Challenge: | Recent advances in context length extension have improved LLMs' performance and safety, but they present critical safety challenges. |
| Approach: | They investigate long-context vulnerabilities in Large Language Models (LLMs) using many-shot jailbreaking to exploit context length extension. |
| Outcome: | The proposed attacks do not require carefully crafted harmful content. |
Margin Matching Preference Optimization: Enhanced Model Alignment with Granular Feedback (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for large language models rely on binary labels that fail to capture the subtle differences in relative quality between pairs. |
| Approach: | They propose a method that incorporates relative quality margins into optimization to improve LLM policies and reward models. |
| Outcome: | The proposed approach outperforms baseline methods on popular benchmarks including MT-bench and RewardBench. |
By My Eyes: Grounding Multimodal Large Language Models with Sensor Data via Visual Prompting (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing text-based prompts for large language models (LLMs) show performance degradation when handling long sensor data sequences. |
| Approach: | They propose a visual prompt that directs MLLMs to utilize visualized sensor data alongside descriptions of the target sensory task. |
| Outcome: | The proposed approach achieves 10% higher accuracy and reduces token costs by 15.8 times on nine sensory tasks involving four sensing modalities . |
Enhancing LLM Agent Safety via Causal Influence Prompting (2025.findings-acl)
Copied to clipboard
| Challenge: | Experimental results demonstrate that our method effectively enhances safety in code execution and mobile device control tasks. |
| Approach: | They propose a technique that leverages causal influence diagrams to identify and mitigate risks arising from agent decision-making. |
| Outcome: | The proposed method enhances safety in code execution and mobile device control tasks. |