Towards Trustworthy LLMs for Code: A Data-Centric Synergistic Auditing Framework

TL;DR

The paper tackles trustworthiness of LLMs for code and the gap between training data quality and evaluation indicators. It proposes DataTrust, a data-centric framework linking evaluation trustworthiness indicators to training data quality indicators and orchestrating Training Data Inspection, Evaluation Data Synthesis, and Root Cause Attribution, plus a Data-Engined Trustworthiness Arena. The framework aims to benefit LLM vendors, evaluators, and users through continuous auditing, leaderboards, and crowd-based input while acknowledging challenges in tool integration, data heterogeneity, and interaction design. If realized, DataTrust could enhance transparency and enable systematic root-cause analyses to accelerate the development of trustworthy LLMs for code.

Abstract

LLM-powered coding and development assistants have become prevalent to programmers' workflows. However, concerns about the trustworthiness of LLMs for code persist despite their widespread use. Much of the existing research focused on either training or evaluation, raising questions about whether stakeholders in training and evaluation align in their understanding of model trustworthiness and whether they can move toward a unified direction. In this paper, we propose a vision for a unified trustworthiness auditing framework, DataTrust, which adopts a data-centric approach that synergistically emphasizes both training and evaluation data and their correlations. DataTrust aims to connect model trustworthiness indicators in evaluation with data quality indicators in training. It autonomously inspects training data and evaluates model trustworthiness using synthesized data, attributing potential causes from specific evaluation data to corresponding training data and refining indicator connections. Additionally, a trustworthiness arena powered by DataTrust will engage crowdsourced input and deliver quantitative outcomes. We outline the benefits that various stakeholders can gain from DataTrust and discuss the challenges and opportunities it presents.

Figures (1)

• Figure 1: Methodology Overview of DataTrust.