Alignment Tuning in Llama3 and DeepSeek-R1 for Adversarial Code Repair Robustness

This report synthesises findings from 9 peer-reviewed papers addressing the following research question: To what extent does alignment tuning in Llama3 and Deepseek R1 mitigate helpfulness degradation across diverse adversarial taxonomies in automated code repair tasks. 8 claims were extracted from source literature; 7 were independently verified against retrieved documents. An automated multi-reviewer quality assessment produced a score of 8.1/10. This report is a machine-generated literature synthesis and does not constitute original research.

Research goal: To what extent does alignment tuning in Llama3 and Deepseek R1 mitigate helpfulness degradation across diverse adversarial taxonomies in automated code repair tasks?

Autonomous literature synthesis. Automated review score: 8.1/10. Full text and citation available at Assignee Research.