提出Chain of Code（CoC）方法，通过让语言模型将语义子任务转为伪代码并选择性模拟解释器执行，扩展了LM的代码推理能力，在BIG-Bench Hard等基准上显著超越Chain of Thought。
摘要：
代码提供了一种通用的语法结构，当与代码解释器结合时，可以构建复杂的程序并执行精确的计算——我们假设语言模型（LMs）可以利用编写代码来改进Chain of Thought推理，不仅适用于逻辑和算术任务，也适用于语义任务（特别是两者混合的任务）。例如，考虑提示LM编写代码来计算它在一篇文章中检测到讽刺的次数：LM可能难以编写一个可由解释器执行的“detect_sarcasm(string)”的实现（处理边缘情况将难以克服）。然而，如果LM不仅编写代码，还通过生成“detect_sarcasm(string)”的预期输出来选择性地“模拟”解释器，LM仍可能产生有效的解决方案。在这项工作中，我们提出了Chain of Code（CoC），这是一种简单但出人意料有效的扩展，可改进LM的代码驱动推理。关键思想是鼓励LM将程序中的语义子任务格式化为灵活的伪代码，解释器可以显式捕获未定义的行为并交给LM（作为“LMulator”）进行模拟。实验表明，Chain of Code在各种基准测试中优于Chain of Thought和其他基线；在BIG-Bench Hard上，Chain of Code达到84%，比Chain of Thought提高了12%。简而言之，CoC通过“用代码思考”拓宽了LM可以回答的推理问题的范围。
英文摘要：
Code provides a general syntactic structure to build complex programs and perform precise computations when paired with a code interpreter - we hypothesize that language models (LMs) can leverage code-writing to improve Chain of Thought reasoning not only for logic and arithmetic tasks, but also for semantic ones (and in particular, those that are a mix of both). For example, consider prompting an LM to write code that counts the number of times it detects sarcasm in an essay: the LM may struggle to write an implementation for "detect_sarcasm(string)" that can be executed by the interpreter (handling the edge cases would be insurmountable). However, LMs may still produce a valid solution if they not only write code, but also selectively "emulate" the interpreter by generating the expected output of "detect_sarcasm(string)". In this work, we propose Chain of Code (CoC), a simple yet surprisingly effective extension that improves LM code-driven reasoning. The key idea is to encourage LMs to format semantic sub-tasks in a program as flexible pseudocode that the interpreter can explicitly catch undefined behaviors and hand off to simulate with an LM (as an "LMulator"). Experiments demonstrate that Chain of Code outperforms Chain of Thought and other baselines across a variety of benchmarks; on BIG-Bench Hard, Chain of Code achieves 84%, a gain of 12% over Chain of Thought. In a nutshell, CoC broadens the scope of reasoning questions that LMs can answer by "thinking in code".
Improving Context Fidelity via Native Retrieval-Augmented Reasoning
DIRO, Université de Montréal（蒙特利尔大学计算机科学与运筹学系）；MetaGPT；Mila - Quebec AI Institute（米拉-魁北克人工智能研究所）；McGill University（麦吉尔大学）；Yale University（耶鲁大学）；Canada CIFAR AI Chair（加拿大CIFAR人工智能教席）
RAG-Star: Enhancing Deliberative Reasoning with Retrieval Augmented Verification and Refinement
Gaoling School of Artificial Intelligence, Renmin University of China（中国人民大学高瓴人工智能学院）；Wuhan University of Science and Technology（武汉科技大学）；National University of Singapore（新加坡国立大学）；Northeastern University at Qinhuangdao（东北大学秦皇岛分校）；BOSS Zhipin（BOSS直聘）
ARise: Towards Knowledge-Augmented Reasoning via Risk-Adaptive Search
Shanghai AI Laboratory（上海人工智能实验室）；Shanghai Innovation Institute（上海创新研究院）；Shanghai Jiao Tong University（上海交通大学）；SenseTime（商汤科技）；Institute of Software, Chinese Academy of Sciences（中国科学院软件研究所）；Tongji University（同济大学）；Hangzhou Institute for Advanced Study（杭州高等研究院）；University of Chinese Academy of Sciences（中国科学院大学）
ChatCoT: Tool-Augmented Chain-of-Thought Reasoning on Chat-based Large Language Models
Gaoling School of Artificial Intelligence, Renmin University of China（中国人民大学高瓴人工智能学院）；School of Information, Renmin University of China（中国人民大学信息学院）；Beijing Key Laboratory of Big Data Management and Analysis Methods（北京大数据管理与分析方法重点实验室）
SciAgent: Tool-augmented Language Models for Scientific Reasoning
Microsoft（微软公司）；University of California, Santa Barbara（加利福尼亚大学圣巴巴拉分校）；Tsinghua University（清华大学）；Singapore Management University（新加坡管理大学）；Nanyang Technological University（南洋理工大学）
Bias-Augmented Consistency Training Reduces Biased Reasoning in Chain-of-Thought
Speechmatics, Apollo Research（Speechmatics与Apollo研究机构）；University of Oxford（牛津大学）；NYU, Anthropic（纽约大学与Anthropic）；NYU（纽约大学）；Anthropic, NYU（Anthropic与纽约大学）
Eliciting Critical Reasoning in Retrieval-Augmented Language Models via Contrastive Explanations
Idiap Research Institute（Idiap研究机构）；Department of Computer Science, University of Manchester（曼彻斯特大学计算机科学系）；National Biomarker Centre (NBC), CRUK Manchester Institute（国家生物标志物中心（NBC），CRUK曼彻斯特研究所）
CoT-RAG: Integrating Chain of Thought and Retrieval-Augmented Generation to Enhance Reasoning in Large Language Models
School of Computer Science and Technology, Huazhong University of Science and Technology, Wuhan, China（华中科技大学计算机科学与技术学院）；Wuhan National Laboratory for Optoelectronics, Huazhong University of Science and Technology, Wuhan, China（华中科技大学光电实验室）；Department of Computer Science, Bowling Green State University, Ohio, USA（Bowling Green State University 计算机科学系）；Department of Computer Science, Aalborg University, Aalborg, Denmark（Aalborg University 计算机科学系）；Huawei Technologies Co., Ltd, Shenzhen, China（华为技术有限公司）
RAGAR, Your Falsehood Radar: RAG-Augmented Reasoning for Political Fact-Checking using Multimodal Large Language Models
Institute for Natural Language Processing, University of Stuttgart（斯图加特大学自然语言处理研究所）；appliedAI Institute for Europe gGmbH（欧洲appliedAI研究所（非营利有限责任公司））；appliedAI Initiative GmbH（appliedAI倡议有限责任公司）
Benchmarking Foundation Models with Retrieval-Augmented Generation in Olympic-Level Physics Problem Solving
AAII, University of Technology Sydney, New South Wales, Australia（AAII，悉尼大学，新南威尔士州，澳大利亚）；Eindhoven University of Technology, Eindhoven, The Netherlands（埃因霍温理工大学，埃因霍温，荷兰）；University of Liverpool, Liverpool, United Kingdom（利物浦大学，利物浦，英国）；University of New South Wales, New South Wales, Australia（新南威尔士大学，新南威尔士州，澳大利亚）
Knowledge Integration Decay in Search-Augmented Reasoning of Large Language Models
Department of Electrical and Computer Engineering, Seoul National University, Seoul, Korea（电气与计算机工程系，首尔国立大学，韩国首尔）；Interdisciplinary Program in Artificial Intelligence, Seoul National University, Seoul, Korea（人工智能交叉学科项目，首尔国立大学，韩国首尔）
Previous studies have shown the efficacy of knowledge augmentation methods in pretrained language models. However, these methods behave differently across domains and downstream tasks. In this work, we investigate the augmentation of pretrained language models with commonsense knowledge in the cause-effect relation classification and commonsense causal reasoning tasks. After automatically verbalizing ATOMIC2020, a wide coverage commonsense reasoning knowledge graph, and GLUCOSE, a dataset of implicit commonsense causal knowledge, we continually pretrain BERT and RoBERTa with the verbalized dat…
Fast, Slow, and Tool-augmented Thinking for LLMs: A Review
School of Mechanical Engineering, Shanghai Jiao Tong University, Shanghai 200240, China（上海交通大学机械工程学院）；School of Computer Science, Shanghai Jiao Tong University, Shanghai 200240, China（上海交通大学计算机科学学院）；Huawei Technologies Co., Ltd., Beijing 100084, China（华为技术有限公司）；Department of Systems Engineering and Engineering Management, The Chinese University of Hong Kong, Hong Kong 999077, China（香港中文大学系统工程与工程管理系）；Department of Computer Science and Engineering, The Chinese University of Hong Kong, Hong Kong 999077, China（香港中文大学计算机科学与工程系）；Huawei Hong Kong Research Center, Hong Kong 999077, China（华为香港研究中心）
Physics Reasoner: Knowledge-Augmented Reasoning for Solving Physics Problems with Large Language Models
Institute for Artificial Intelligence, Tsinghua University (THUAI)（清华大学人工智能研究院）；Beijing National Research Center for Information Science and Technology (BNRist)（北京信息科学与技术国家研究中心）；Department of Automation, Tsinghua University（清华大学自动化系）；TMLR Group, Hong Kong Baptist University（香港浸会大学TMLR组）
ReARTeR: Retrieval-Augmented Reasoning with Trustworthy Process Rewarding
Kuaishou Technology Co., Ltd.（快手科技有限公司）；Institute for Clarity in Documentation（文献清晰度研究所）；Inria Paris-Rocquencourt（法国国家信息与自动化研究所巴黎-罗康分所）；Rajiv Gandhi University（拉吉夫甘地大学）；Tsinghua University（清华大学）；Palmer Research Laboratories（帕尔默研究实验室）
