Measuring Progress on Scalable Oversight for Large Language Models
Anthropic（Anthropic公司）；Surge AI（Surge AI公司）
Samuel R. Bowman, Jeeyoon Hyun, Ethan Perez, Edwin Chen, Craig Pettit, Scott Heiner, Kamilė Lukošiūtė, Amanda Askell, Andy Jones, Anna Chen, Anna Goldie, Azalia… 展开作者
Samuel R. Bowman, Jeeyoon Hyun, Ethan Perez, Edwin Chen, Craig Pettit, Scott Heiner, Kamilė Lukošiūtė, Amanda Askell, Andy Jones, Anna Chen, Anna Goldie, Azalia Mirhoseini, Cameron McKinnon, Christopher Olah, Daniela Amodei, Dario Amodei, Dawn Drain, Dustin Li, Eli Tran-Johnson, Jackson Kernion, Jamie Kerr, Jared Mueller, Jeffrey Ladish, Joshua Landau, Kamal Ndousse, Liane Lovitt, Nelson Elhage, Nicholas Schiefer, Nicholas Joseph, Noemí Mercado, Nova DasSarma, Robin Larson, Sam McCandlish, Sandipan Kundu, Scott Johnston, Shauna Kravec, Sheer El Showk, Stanislav Fort, Timothy Telleen-Lawton, Tom Brown, Tom Henighan, Tristan Hume, Yuntao Bai, Zac Hatfield-Dodds, Ben Mann, Jared Kaplan
Developing safe and useful general-purpose AI systems will require us to make progress on scalable oversight: the problem of supervising systems that potentially outperform us on most skills relevant to the task at hand. Empirical work on this problem is not straightforward, since we do not yet have systems that broadly exceed our abilities. This paper discusses one of the major ways we think about this problem, with a focus on ways it can be studied empirically. We first present an experimental design centered on tasks for which human specialists succeed but unaided humans and current general AI systems fail. We then present a proof-of-concept experiment meant to demonstrate a key feature of this experimental design and show its viability with two question-answering tasks: MMLU and time-limited QuALITY. On these tasks, we find that human participants who interact with an unreliable large-language-model dialog assistant through chat -- a trivial baseline strategy for scalable oversight -- substantially outperform both the model alone and their own unaided performance. These results are an encouraging sign that scalable oversight will be tractable to study with present models and bolster recent findings that large language models can productively assist humans with difficult tasks.
Is Multihop QA in DiRe Condition? Measuring and Reducing Disconnected Reasoning
Stony Brook University（石溪大学）；Allen Institute for AI（艾伦人工智能研究所）
在多跳问答方面有真正的进展吗？模型通常利用数据集工件来产生正确的答案，而不将多个支持事实的信息连接起来。这限制了我们衡量真实进展的能力，并违背了构建多跳QA数据集的目的。我们为解决这一问题作出三项贡献。首先，我们将这种不受欢迎的行为形式化为跨支持事实子集的断开推理。这允许开发一个模型不可知的探针，用于测量任何模型可以通过断开推理欺骗多少。其次，使用\emph{contrastive support sufficiency}的概念，我们引入了现有数据集的自动转换，减少了断开推理的数量。第三，我们的实验表明，在阅读理解环境中，多跳问答并没有取得太大的进展。对于最近的一个大规模模型（XLNet），我们发现，在HotpotQA上的F1得分72中，只有18分是通过多事实推理获得的，与简单的RNN基线大致相同。我们的转换大大减少了断开推理（答案F1中的19分）。它是对抗性办法的补充，共同产生进一步的削减。
AgentMisalignment: Measuring the Propensity for Misaligned Behaviour in LLM-Based Agents
Department of Computer Science（计算机科学系）；University of Oxford（牛津大学）；Institute of Intelligent Systems and Robotics（智能系统与机器人研究所）；Sorbonne Université（索邦大学）；The Leverhulme Centre for the Future of Intelligence（未来智能中心）；University of Cambridge（剑桥大学）；Independent Researcher（独立研究者）；Department of Computer Science and Technology（计算机科学与技术系）；Department of Engineering（工程系）
BAITBENCH: Measuring Agent Reward Hacking with Optional Shortcuts Planted in ML Tasks
National University of Singapore（新加坡国立大学）；MIT（麻省理工学院）；Vrije Universiteit Amsterdam（阿姆斯特丹自由大学）；University of Toronto（多伦多大学）；University of Wisconsin-Madison（威斯康星大学麦迪逊分校）；Arb Research（Arb研究机构）
Measuring Human Involvement in AI-Generated Text: A Case Study on Academic Writing
Dept. of Information & Communication Engineering (ICE), The University of Tokyo (Japan)（信息与通信工程系（ICE），东京大学（日本））；National Institute of Informatics (Japan)（日本信息处理学会）
Large Language Models Illuminate a Progressive Pathway to Artificial Healthcare Assistant: A Review
Peking University（北京大学）；Peking University Cancer Hospital and Institute（北京大学肿瘤医院暨北京市肿瘤防治研究所）；National Engineering Laboratory for Big Data Analysis and Applications, Peking University（北京大学大数据分析与应用技术国家工程实验室）；Peking University Changsha Institute for Computing and Digital Economy（北京大学长沙计算与数字经济研究院）
Towards Trustworthy AI: A Review of Ethical and Robust Large Language Models
Canizaro Livingston Gulf States Center for Environmental Informatics, the University of New Orleans（新奥尔良大学）；Center for Geospatial Sciences, Naval Research Laboratory（海军研究实验室）；US Army Corps of Engineers, Engineer Research and Development Center（美国陆军工兵部工程师研究与发展中心）
What and Whose Knowledge? Measuring Epistemic Diversity in Large Language Models
University of Copenhagen（哥本哈根大学）；Aalborg University Copenhagen（奥胡斯大学哥本哈根分校）；Stanford University（斯坦福大学）；University of Colorado Boulder（科罗拉多大学波德分校）；Microsoft Research（微软研究院）
