Search for: All records

Recent advances in large language models (LLMs) have transformed software development by automatically generating code from natural language. Yet challenges remain in generating fully correct code that aligns with user intent. Our study reveals that LLMs tend to pay less attention to user prompts as more code tokens are generated. We hypothesize that this attention dilution issue is an important reason for code generation errors. To mitigate this issue, we propose Selective Prompt Anchoring (SPA) to guide code LLMs to pay more attention to user intent when generating code. We evaluate SPA using six base LLMs across six benchmarks. Our results demonstrate that SPA enhances Pass@1 by up to 12.9%, consistently outperforming SOTA methods in all settings. Our code is available at https://github.com/magic-YuanTian/Selective- Prompt-Anchoring. 

Autonomous driving systems (ADS) require extensive testing and validation before deployment. However, it is tedious and time-consuming to construct traffic scenarios for ADS testing. In this paper, we propose TrafficComposer, a multi-modal traffic scenario construction approach for ADS testing. TrafficComposer takes as input a natural language (NL) description of a desired traffic scenario and a complementary traffic scene image. Then, it generates the corresponding traffic scenario in a simulator, such as CARLA and LGSVL. Specifically, TrafficComposer integrates high-level dynamic information about the traffic scenario from the NL description and intricate details about the surrounding vehicles, pedestrians, and the road network from the image. The information from the two modalities is complementary to each other and helps generate high-quality traffic scenarios for ADS testing. On a benchmark of 120 traffic scenarios, TrafficComposer achieves 97.0% accuracy, outperforming the best-performing baseline by 7.3%. Both direct testing and fuzz testing experiments on six ADSs prove the bug detection capabilities of the traffic scenarios generated by TrafficComposer. These scenarios can directly discover 37 bugs and help two fuzzing methods find 33%–124% more bugs serving as initial seeds. 

Large Language Models (LLMs) have demonstrated unprecedented capability in code generation. However, LLM-generated code is still plagued with a wide range of functional errors, especially for complex programming tasks that LLMs have not seen before. Recent studies have shown that developers often struggle with inspecting and fixing incorrect code generated by LLMs, diminishing their productivity and trust in LLM-based code generation. Inspired by the mutual grounding theory in communication, we propose an interactive approach that leverages code comments as a medium for developers and LLMs to establish a shared understanding. Our approach facilitates iterative grounding by interleaving code generation, inline comment generation, and contextualized user feedback through editable comments to align generated code with developer intent. We evaluated our approach on two popular benchmarks and demonstrated that our approach significantly improved multiple state-of-the-art LLMs, e.g., 17.1% pass@1 improvement for code-davinci-002 on HumanEval. Furthermore, we conducted a user study with 12 participants in comparison to two baselines: (1) interacting with GitHub Copilot, and (2) interacting with a multi-step code generation paradigm called Multi-Turn Program Synthesis. Participants completed the given programming tasks 16.7% faster and with 10.5% improvement in task success rate when using our approach. Both results show that interactively refining code comments enables the collaborative establishment of mutual grounding, leading to more accurate code generation and higher developer confidence. 

Large Language Models (LLMs) have demonstrated unprecedented capabilities in code generation. However, there remains a limited understanding of code generation errors that LLMs can produce. To bridge the gap, we conducted an in-depth analysis of code generation errors across six representative LLMs on the HumanEval dataset. Specifically, we first employed open coding and thematic analysis to distill a comprehensive taxonomy of code generation errors. We analyzed two dimensions of error characteristics -- semantic characteristics and syntactic characteristics. Our analysis revealed that LLMs often made non-trivial, multi-line code generation errors in various locations and with various root causes. We further analyzed the correlation between these errors and task complexity as well as test pass rate. Our findings highlighted several challenges in locating and fixing code generation errors made by LLMs. In the end, we discussed several future directions to address these challenges. 

This paper introduces a novel sampling technique based on the dynamics of a 2-state Quantum Walk (QW) in a one-dimensional space. By leveraging concepts from nonparametric statistics, specifically the kernel smoothing method, our approach addresses two key challenges in Quantum Walk sampling: discontinuities in sampling distributions and potential inaccuracies in limiting distributions. Our innovative method effectively mitigates these issues, leading to significant improvements in density estimation and sampling efficacy compared to traditional Quantum Walk distributions and sampling techniques.