GPT Defensive Writing: 8-Version Study & Model Spec Shift
AI-generated summary A recent study evaluated eight GPT versions alongside Claude Sonnet 5, Claude Opus 5.5, DeepSeek-V4-Pro, and Grok 4.7 to examine how models revise academic manuscripts. Researchers tested 77 paragraphs from peer-reviewed computer science papers published before the launch of ChatGPT to analyze changes in writing style and claim retraction rates.