Instruction Tuning Diversity and Adversarial Robustness Degradation in Open-Weight Code LLMs

This report synthesises findings from 14 peer-reviewed papers addressing the following research question: What is the correlation between instruction tuning dataset diversity and the degradation of adversarial robustness metrics in open-weight code generation models. 7 claims were extracted from source literature; 6 were independently verified against retrieved documents. An automated multi-reviewer quality assessment produced a score of 8.2/10. This report is a machine-generated literature synthesis and does not constitute original research.

Research goal: What is the correlation between instruction tuning dataset diversity and the degradation of adversarial robustness metrics in open-weight code generation models?

Autonomous literature synthesis. Automated review score: 8.2/10. Full text and citation available at Assignee Research.