Black Box LLM

Black-box large language models (LLMs), characterized by their opaque internal workings, are a focus of intense research aimed at understanding their vulnerabilities and improving their safety and reliability. Current research explores methods for adversarial attacks (e.g., "jailbreaking" through prompt manipulation), optimization of token usage for efficiency, and techniques for evaluating and improving model robustness and alignment with human values, often employing reinforcement learning and iterative distillation methods. These investigations are crucial for mitigating risks associated with deploying LLMs in real-world applications and for advancing the development of more trustworthy and beneficial AI systems.

Papers

October 28, 2024

Matryoshka: Learning to Drive Black-Box LLMs with LLMs
Changhao Li, Yuchen Zhuang, Rushi Qiang, Haotian Sun, Hanjun Dai, Chao Zhang, Bo Dai
Medical LLM Controllable Generation Black Box LLM

October 2, 2024

FlipAttack: Jailbreak LLMs via Flipping
Yue Liu, Xiaoxin He, Miao Xiong, Jinlan Fu, Shumin Deng, Bryan Hooi
Jailbreak Attack Attack Success Rate Motor Augmented Mode Black Box LLM Negative Flip Flip Attack

October 1, 2024

Optimizing Token Usage on Large Language Model Conversations Using the Design Structure Matrix
Ramon Maria Garcia Alarcia, Alessandro Golkar
Conversational Large Language Model Black Box LLM Conversational LLM Token Optimization Structural Design

August 11, 2024

Kov: Transferable and Naturalistic Black-Box LLM Attacks using Markov Decision Processes and Tree Search
Robert J. Moss
Adversarial Attack Markov Decision Process Tree Search Black Box LLM Word Level Attack LLM Attack Token Level Attack

August 6, 2024

Compromising Embodied Agents with Contextual Backdoor Attacks
Aishan Liu, Yuguang Zhou, Xianglong Liu, Tianyuan Zhang, Siyuan Liang, Jiakai Wang, Yanjun Pu, Tianlin Li, Junqi Zhang, Wenbo Zhou, Qing Guo, Dacheng Tao
Embodied Agent Context Dependent Black Box LLM Defect Sample Domain Demonstration Adversarial Context

May 24, 2024

A Solution-based LLM API-using Methodology for Academic Information Seeking
Yuanchun Wang, Jifan Yu, Zijun Yao, Jing Zhang, Yuyang Xie, Shangqing Tu, Yiyang Fu, Youhe Feng, Jinkai Zhang, Jingyao Zhang, Bowen Huang, Yuanyao Li, Huihui Yuan, Lei Hou, Juanzi Li, Jie Tang
Large Language Model NCD Method LLM Based State of the Art LLM Scholarly Document Black Box LLM

April 30, 2024

Harmonic LLMs are Trustworthy
Nicholas S. Kersting, Mohammad Rahman, Suchismitha Vedala, Yang Wang
Native Robustness High Explainability Adversarial Prompt Black Box LLM Annotation Study Different LLM

April 26, 2024

Reinforcement Retrieval Leveraging Fine-grained Feedback for Fact Checking News Claims with Black-Box LLM
Xuan Zhang, Wei Gao
Fine Grained Fact Checking Retrieval Augmented Language Model Advanced Large Language Model Black Box LLM Fine Grained Feedback

April 21, 2024

AdvPrompter: Fast Adaptive Adversarial Prompting for LLMs
Anselm Paulus, Arman Zharmagambetov, Chuan Guo, Brandon Amos, Yuandong Tian
Large Language Model Adversarial Prompt Adversarial Suffix Black Box LLM Adversarial Prompting

March 14, 2024

Logits of API-Protected LLMs Leak Proprietary Information
Matthew Finlayson, Xiang Ren, Swabha Swayamdipta
Large Language Model API Usage Second Ranked Logits LLM Output Black Box LLM

January 18, 2024

Sketch-Guided Constrained Decoding for Boosting Blackbox Large Language Models without Logit Access
Saibo Geng, Berkay Döner, Chris Wendler, Martin Josifoski, Robert West
Large Language Model Text Generation Second Ranked Logits Black Box Large Language Model Constrained Decoding Language Model Output Black Box LLM

January 6, 2024

Malla: Demystifying Real-world Large Language Model Integrated Malicious Services
Zilong Lin, Jian Cui, Xiaojing Liao, XiaoFeng Wang
Language Model Real World eXploitation Black Box LLM Threat Landscape

November 18, 2023

(Why) Is My Prompt Getting Worse? Rethinking Regression Testing for Evolving LLM APIs
Wanqin Ma, Chenyang Yang, Christian Kästner
Language Model Style PROMPT Black Box LLM Property Based

October 25, 2023

CycleAlign: Iterative Distillation from Black-box LLM to White-box Models for Better Human Alignment
Jixiang Hong, Quan Tu, Changyu Chen, Xing Gao, Ji Zhang, Rui Yan
Large Language Model Language Model White Box Model Generated Human Alignment Black Box LLM

June 22, 2023

Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMs
Miao Xiong, Zhiyuan Hu, Xinyang Lu, Yifei Li, Jie Fu, Junxian He, Bryan Hooi
Large Language Model Medical LLM High Uncertainty Anticipation Empirical Evaluation Confidence Calibration Black Box LLM LLM Uncertainty Confidence Elicitation

June 12, 2023

TrojLLM: A Black-box Trojan Prompt Attack on Large Language Models
Jiaqi Xue, Mengxin Zheng, Ting Hua, Yilin Shen, Yepeng Liu, Ladislau Boloni, Qian Lou
Trojan Attack Black Box LLM

June 5, 2023

InstructZero: Efficient Instruction Optimization for Black-Box Large Language Models
Lichang Chen, Jiuhai Chen, Tom Goldstein, Heng Huang, Tianyi Zhou
Black Box Large Language Model Black Box LLM Instruction Optimization Prompt Vector

May 25, 2023

On the Tool Manipulation Capability of Open-source Large Language Models
Qiantong Xu, Fenglu Hong, Bo Li, Changran Hu, Zhengyu Chen, Jian Zhang
Large Language Model Open Source Large Language Model Closed Source Model Black Box LLM Tool Use Tool Manipulation

May 9, 2023

FrugalGPT: How to Use Large Language Models While Reducing Cost and Improving Performance
Lingjiao Chen, Matei Zaharia, James Zou
Large Language Model System Performance Hidden CoST Inference Cost Adaptive Prompt Black Box LLM LLM Accuracy

March 31, 2023

Evaluating GPT-4 and ChatGPT on Japanese Medical Licensing Examinations
Jungo Kasai, Yuhei Kasai, Keisuke Sakaguchi, Yutaro Yamada, Dragomir Radev
Large Language Model ChatGPT Generated Conversation GPT 4 Japanese Text State Medical Licensing Examination Black Box LLM

Black Box LLM

Papers

Matryoshka: Learning to Drive Black-Box LLMs with LLMs

FlipAttack: Jailbreak LLMs via Flipping

Optimizing Token Usage on Large Language Model Conversations Using the Design Structure Matrix

Kov: Transferable and Naturalistic Black-Box LLM Attacks using Markov Decision Processes and Tree Search

Compromising Embodied Agents with Contextual Backdoor Attacks

A Solution-based LLM API-using Methodology for Academic Information Seeking

Harmonic LLMs are Trustworthy

Reinforcement Retrieval Leveraging Fine-grained Feedback for Fact Checking News Claims with Black-Box LLM

AdvPrompter: Fast Adaptive Adversarial Prompting for LLMs

Logits of API-Protected LLMs Leak Proprietary Information

Sketch-Guided Constrained Decoding for Boosting Blackbox Large Language Models without Logit Access

Malla: Demystifying Real-world Large Language Model Integrated Malicious Services

(Why) Is My Prompt Getting Worse? Rethinking Regression Testing for Evolving LLM APIs

CycleAlign: Iterative Distillation from Black-box LLM to White-box Models for Better Human Alignment

Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMs

TrojLLM: A Black-box Trojan Prompt Attack on Large Language Models

InstructZero: Efficient Instruction Optimization for Black-Box Large Language Models

On the Tool Manipulation Capability of Open-source Large Language Models

FrugalGPT: How to Use Large Language Models While Reducing Cost and Improving Performance

Evaluating GPT-4 and ChatGPT on Japanese Medical Licensing Examinations