Large Language Models as Robust Data Generators in Software Analytics: Are We There Yet?
Bibliographic record
Abstract
Large Language Model (LLM)-generated data is increasingly used in software analytics, but it is unclear how this data compares to human-written data, particularly when models are exposed to adversarial scenarios. Adversarial attacks can compromise the reliability and security of software systems, so understanding how LLM-generated data performs under these conditions, compared to human-written data, which serves as the benchmark for model performance, can provide valuable insights into whether LLM-generated data offers similar robustness and effectiveness. To address this gap, we systematically evaluate and compare the quality of human-written and LLM-generated data for fine-tuning robust pre-trained models (PTMs) in the context of adversarial attacks. We evaluate the robustness of six widely used PTMs, fine-tuned on human-written and LLM-generated data, before and after adversarial attacks. This evaluation employs nine state-of-the-art (SOTA) adversarial attack techniques across three popular software analytics tasks: clone detection, code summarization, and sentiment analysis in code review discussions. Additionally, we analyze the quality of the generated adversarial examples using eleven similarity metrics. Our findings reveal that while PTMs fine-tuned on LLM-generated data perform competitively with those fine-tuned on human-written data, they exhibit less robustness against adversarial attacks in software analytics tasks. Our study underscores the need for further exploration into enhancing the quality of LLM-generated training data to develop models that are both high-performing and capable of withstanding adversarial attacks in software analytics.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.001 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.001 | 0.003 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.007 | 0.009 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; both teacher heads agree on what is shown here.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".