OOPSLA
OOPSLA 2024
Drowzee: Metamorphic Testing for Fact-Conflicting Hallucination Detection in Large Language Models
Abstract: Large language models (LLMs) have revolutionized language processing, but face critical challenges with security, privacy, and generating hallucinations — coherent but factually inaccurate outputs. A major issue is fact-conflicting hallucination (FCH), where LLMs produce content contradicting ground truth facts. Addressing FCH is difficult due to two key challenges: 1) Automatically constructing and updating benchmark datasets is hard, as existing methods rely on manually curated static benchmarks that cannot cover the broad, evolving spectrum of FCH cases. 2) Validating the reasoning behind LLM outputs is inherently difficult, especially for complex logical relations. To tackle these challenges, we introduce a novel logic-programming-aided metamorphic testing technique for FCH detection. We develop an extensive and extensible framework that constructs a comprehensive factual knowledge base by crawling sources like Wikipedia, seamlessly integrated into Drowzee. Using logical reasoning rules, we transform and augment this knowledge into a large set of test cases with ground truth answers. We test LLMs on these cases through template-based prompts, requiring them to provide reasoned answers. To validate their reasoning, we propose two semantic-aware oracles that assess the similarity between the semantic structures of the LLM answers and ground truth. Our approach automatically generates useful test cases and identifies hallucinations across six LLMs within nine domains, with hallucination rates ranging from 24.7% to 59.8%. Key findings include LLMs struggling with temporal concepts, out-of-distribution knowledge, and lack of logical reasoning capabilities. The results show that logic-based test cases generated by Drowzee effectively trigger and detect hallucinations. To further mitigate the identified FCHs, we explored model editing techniques, which proved effective on a small scale (with edits to fewer than 1000 knowledge pieces). Our findings emphasize the need for continued community efforts to detect and mitigate model hallucinations.
OOPSLA 2017
Automated testing of graphics shader compilers
Abstract: We present an automated technique for finding defects in compilers for graphics shading languages. key challenge in compiler testing is the lack of an oracle that classifies an output as correct or incorrect; this is particularly pertinent in graphics shader compilers where the output is a rendered image that is typically under-specified. Our method builds on recent successful techniques for compiler validation based on metamorphic testing, and leverages existing high-value graphics shaders to create sets of transformed shaders that should be semantically equivalent. Rendering mismatches are then indicative of shader compilation bugs. Deviant shaders are automatically minimized to identify, in each case, a minimal change to an original high-value shader that induces a shader compiler bug. We have implemented the approach as a tool, GLFuzz, targeting the OpenGL shading language, GLSL. Our experiments over a set of 17 GPU and driver configurations, spanning the main 7 GPU designers, have led to us finding and reporting more than 60 distinct bugs, covering all tested configurations. As well as defective rendering, these issues identify security-critical vulnerabilities that affect WebGL, including a significant remote information leak security bug where a malicious web page can capture the contents of other browser tabs, and a bug whereby visiting a malicious web page can lead to a ``blue screen of death’’ under Windows 10. Our findings show that shader compiler defects are prevalent, and that metamorphic testing provides an effective means for detecting them automatically.