Reinforcement learning from human feedback (RLHF) is fundamentally limited by the capacity of humans to correctly evaluate model output. To improve human evaluation ability and overcome that limitation this work trains "critic" models that help humans to more accurately evaluate model-written code. These critics are themselves LLMs trained with RLHF to write natural language feedback highlighting problems in code from real-world assistant tasks. On code containing naturally occurring LLM errors model-written critiques are preferred over human critiques in 63% of cases, and human evaluation finds that models catch more bugs than human contractors paid for code review. We further confirm that our fine-tuned LLM critics can successfully identify hundreds of errors in ChatGPT training data rated as "flawless", even though the majority of those tasks are non-code tasks and thus out-of-distribution for the critic model. Critics can have limitations of their own, including hallucinated bugs that could mislead humans into making mistakes they might have otherwise avoided, but human-machine teams of critics and contractors catch similar numbers of bugs to LLM critics while hallucinating less than LLMs alone.
School of Computer Science, Peking University(北京大学计算机学院);Institute for Artificial Intelligence, Peking University(北京大学人工智能研究院);School of Software, Tsinghua University(清华大学软件学院)
Rise From The Ashes: LLM-based Static Analysis for Deep Learning Framework Bugs
State Key Laboratory for Novel Software Technology, Nanjing University(南京大学软件新技术国家重点实验室);School of Artificial Intelligence and Computer Science, Nantong University(南通大学人工智能与计算机学院);College of Computer Science and Technology/College of Software, Nanjing University of Aeronautics and Astronautics(南京航空航天大学计算机科学与技术学院/软件学院);Beijing Academy of Artificial Intelligence(北京智源人工智能研究院)
Practical Testing of a C99 Compiler Using Output Comparison
本文提出了一种测试C99编译器的简单方法,通过将其输出与已有工具的输出进行比较。这种方法的优点是可以从现有的源代码中批量添加新的测试用例,减少了对正确性问题的深入调查以及手工创建新测试代码的需要。这项技术被用于测试PalmSource Palm OS Cobalt ARM C/C++交叉编译器,主要用于测试Palm-Powered个人数字助理的标准合规性和生成代码的正确执行。这里描述的技术发现了数百个bug,大部分在我们的内部代码中,但也在Edison Design Group和Apogee Software长期以来的高质量前端和后端代码中。它还在GNU C编译器中发现了18个bug,以及一个特定于Apple版本的GCC的bug,一个特定于Suse版本的GCC的bug,以及ARM处理器的GCC版本中的十几个bug,其中几个是关键的。
FeatureFuzz: Treating Bug-prone Features as Core Primitives for Compiler Fuzzing
State Key Laboratory of Novel Software Technology, Nanjing University(南京大学软件新技术国家重点实验室);Institute of Dataspace, Hefei Comprehensive National Science Center(合肥综合性国家科学中心数据空间研究所);Honor Device Co., Ltd(荣耀终端有限公司)