Comment Comment by cowlevel Chinese models also wrote code with 23% more vulnerabilities if told they're writing for the DoD than for the Chinese equivalent. parent cowlevel 8 days ago
