> GPT-4V can easily whitewash its entire copyrighted training corpus to be unrecognizably distinct Is that just by increasing the temperature, tweaking the pro

> GPT-4V can easily whitewash its entire copyrighted training corpus to be unrecognizably distinct Is that just by increasing the temperature, tweaking the prompt, etc.? If you can operate on the raw weights and recreate the original text, copyright infringement still applies.