Nicholas Carlini, Florian Tramer, Eric Wallace, Matthew Jagielski, Ariel Herbert-Voss, Katherine Lee, Adam Roberts, Tom Brown, Dawn Song, Ulfar Erlingsson, Alin… 展开作者
Nicholas Carlini, Florian Tramer, Eric Wallace, Matthew Jagielski, Ariel Herbert-Voss, Katherine Lee, Adam Roberts, Tom Brown, Dawn Song, Ulfar Erlingsson, Alina Oprea, Colin Raffel
It has become common to publish large (billion parameter) language models that have been trained on private datasets. This paper demonstrates that in such settings, an adversary can perform a training data extraction attack to recover individual training examples by querying the language model.
We demonstrate our attack on GPT-2, a language model trained on scrapes of the public Internet, and are able to extract hundreds of verbatim text sequences from the model's training data. These extracted examples include (public) personally identifiable information (names, phone numbers, and email addresses), IRC conversations, code, and 128-bit UUIDs. Our attack is possible even though each of the above sequences are included in just one document in the training data.
We comprehensively evaluate our extraction attack to understand the factors that contribute to its success. Worryingly, we find that larger models are more vulnerable than smaller models. We conclude by drawing lessons and discussing possible safeguards for training large language models.
Allen Institute for Artificial Intelligence（艾伦人工智能研究所）；Arrive Bio（Arrive Bio（阿里夫生物））
Prior work on controllable text generation has focused on learning how to control language models through trainable decoding, smart-prompt design, or fine-tuning based on a desired objective. We hypothesize that the information needed to steer the model to generate a target sentence is already encoded within the model. Accordingly, we explore a different approach altogether: extracting latent vectors directly from pretrained language model decoders without fine-tuning. Experiments show that there exist steering vectors, which, when added to the hidden states of the language model, generate a t…
Can world knowledge learned by large language models (LLMs) be used to act in interactive environments? In this paper, we investigate the possibility of grounding high-level tasks, expressed in natural language (e.g. "make breakfast"), to a chosen set of actionable steps (e.g. "open fridge"). While prior work focused on learning from explicit step-by-step examples of how to act, we surprisingly find that if pre-trained LMs are large enough and prompted appropriately, they can effectively decompose high-level tasks into mid-level plans without any further training. However, the plans produced n…
MasonNLP+ at SemEval-2023 Task 8: Extracting Medical Questions, Experiences and Claims from Social Media using Knowledge-Augmented Pre-trained Language Models
Recent work (e.g. LAMA (Petroni et al., 2019)) has found that the quality of the factual information extracted from Large Language Models (LLMs) depends on the prompts used to query them. This inconsistency is problematic because different users will query LLMs for the same information using different wording, but should receive the same, accurate responses regardless. In this work we aim to address this shortcoming by introducing P-Adapters: lightweight models that sit between the embedding layer and first attention layer of LLMs. They take LLM embeddings as input and output continuous prompt…
AI-Enhanced Cognitive Behavioral Therapy: Deep Learning and Large Language Models for Extracting Cognitive Pathways from Social Media Texts
School of Software Engineering, Beijing University of Technology（北京工业大学软件学院）；School of Nursing, Wuhan University（武汉大学护理学院）；Sorbonne Université, Institut du Cerveau – Paris Brain Institute - ICM, CNRS, Inria, Inserm, AP-HP, Hôpital de la Pitié-Salpêtrière（索邦大学，巴黎脑研究所-ICM，法国国家科学研究中心，法国国家信息与自动化研究所，法国国家健康与医学研究院，巴黎公共医疗集团，皮蒂-萨尔皮特里埃医院）
Semantic 3D scene understanding is a problem of critical importance in robotics. While significant advances have been made in simultaneous localization and mapping algorithms, robots are still far from having the common sense knowledge about household objects and their locations of an average human. We introduce a novel method for leveraging common sense embedded within large language models for labelling rooms given the objects contained within. This algorithm has the added benefits of (i) requiring no task-specific pre-training (operating entirely in the zero-shot regime) and (ii) generalizi…
Large Language Model for Extracting Complex Contract Information in Industrial Scenes
Shanghai Research Institute for Intelligent Autonomous Systems, Tongji University, Shanghai, China（上海智能自主系统研究院，同济大学）；School of Computer Science and Technology, East China Normal University, Shanghai, China（计算机科学与技术学院，华东师范大学）
Can large language models replace humans in the systematic review process? Evaluating GPT-4's efficacy in screening and extracting data from peer-reviewed and grey literature in multiple languages
Trinity Centre for Global Health（三一全球健康中心）；School of Psychology（心理学院）；Trinity College Dublin（都柏林圣三一学院）；University of York（约克大学）
A Multi-Stage Large Language Model Framework for Extracting Suicide-Related Social Determinants of Health
Cockrell School of Engineering, The University of Texas at Austin（德克萨斯大学奥斯汀分校工程学院）；Population Health Sciences, Weill Cornell Medicine（韦尔·柯尔医学中心流行病学与卫生科学系）；School of Information, The University of Texas at Austin（德克萨斯大学奥斯汀分校信息学院）；Department of Biomedical Informatics, Columbia University（哥伦比亚大学生物医学信息学系）
Many efforts have been devoted to extracting constituency trees from pre-trained language models, often proceeding in two stages: feature definition and parsing. However, this kind of methods may suffer from the branching bias issue, which will inflate the performances on languages with the same branch it biases to. In this work, we propose quantitatively measuring the branching bias by comparing the performance gap on a language and its reversed language, which is agnostic to both language models and extracting methods. Furthermore, we analyze the impacts of three factors on the branching bia…
Extracting and Combining Abilities For Building Multi-lingual Ability-enhanced Large Language Models
Gaoling School of Artificial Intelligence, Renmin University of China（中国人民大学人工智能学院）；School of Information, Renmin University of China（中国人民大学信息学院）；Baichuan Inc（百川科技）
Extracting Patient History from Clinical Text: A Comparative Study of Clinical Large Language Models
Oklahoma State University（俄克拉荷马州立大学）；University of South Florida（南佛罗里达大学）；South Dakota School of Mines and Technology（南达科他矿业理工学院）；University of South Florida Morsani College of Medicine（南佛罗里达大学莫萨尼医学院）；South University School of Pharmacy（南大学药学院）；The State University of New York Upstate Medical University（纽约州立大学上州医科大学）；The State University of New York at New Paltz（纽约州立大学新帕尔茨分校）
Extracting and Encoding: Leveraging Large Language Models and Medical Knowledge to Enhance Radiological Text Representation
Pontificia Universidad Católica de Chile（智利天主教大学）；University of Pennsylvania（宾夕法尼亚大学）；KU Leuven（鲁汶大学）；Millennium Institute for Intelligent Healthcare Engineering (iHEALTH)（千禧智能医疗工程研究院（iHEALTH））；National Center for Artificial Intelligence (CENIA)（国家人工智能中心（CENIA））
Extracting effective solutions hidden in large language models via generated comprehensive specialists: case studies in developing electronic devices
School of Science, Yokohama City University（横滨市立大学理学院）；Institute of Science Tokyo（东京科学大学）；MDX Research Center for Element Strategy（MDX元素战略研究中心）；Materials Integration Laboratories, AGC Inc（AGC株式会社材料集成实验室）；Graduate School of Medical Life Science, Yokohama City University（横滨市立大学医生命科学研究生院）
Extracting Social Support and Social Isolation Information from Clinical Psychiatry Notes: Comparing a Rule-based NLP System and a Large Language Model
Weill Cornell Medicine（威尔康奈尔医学院）；Icahn School of Medicine at Mount Sinai（西奈山伊坎医学院）；Mayo Clinic（梅奥诊所）；Columbia University Vagelos College of Physicians and Surgeons（哥伦比亚大学瓦格洛斯内科医生与外科医生学院）；New York State Psychiatric Institute（纽约州精神病学研究所）
EMRModel: A Large Language Model for Extracting Medical Consultation Dialogues into Structured Medical Records
Taihe County People’s Hospital（太和县人民医院）；Innovation + Research Institute（创新+研究院）；GuoChuang Cloud Technology Ltd（国创云科技有限公司）；Xunfei Healthcare Technology Ltd（讯飞医疗科技有限公司）；University of Southern California（南加州大学）；The First Affiliated Hospital of Anhui Medical University（安徽医科大学第一附属医院）
Extracting Consumer Insight from Text: A Large Language Model Approach to Emotion and Evaluation Measurement
Monash University（莫纳什大学）；University of New South Wales（新南威尔士大学）；Flinders University（弗林德斯大学）；Babson College（巴布森学院）；University of Bath（巴斯大学）；Tecnológico de Monterrey（蒙特雷理工学院）
A Path Towards Legal Autonomy: An interoperable and explainable approach to extracting, transforming, loading and computing legal information using large language models, expert systems and Bayesian networks
The University of Sussex（萨塞克斯大学）；Maastricht University（马斯特里赫特大学）；Massachusetts Institute of Technology（麻省理工学院）；VERSES Research Lab（VERSES 研究实验室）；Macquarie University（麦考瑞大学）；Université du Québec à Montréal（蒙特利尔魁北克大学）；University College London（伦敦大学学院）
Extracting Social Determinants of Health from Pediatric Patient Notes Using Large Language Models: Novel Corpus and Methods
University of Washington（华盛顿大学）；George Mason University（乔治梅森大学）；Dana-Farber Cancer Institute（达纳-法伯癌症研究所）；Boston Children's Hospital（波士顿儿童医院）；Harvard Medical School（哈佛医学院）
Simple Yet Effective: Extracting Private Data Across Clients in Federated Fine-Tuning of Large Language Models
Harbin Institute of Technology（哈尔滨工业大学）；Kuaishou Technology（快手科技）；Meta AI；Monash University（墨尔本大学）；Fudan University（复旦大学）；Shanghai Academy of Artificial Intelligence for Science（上海人工智能科学研究院）
