https://www.landofgpt.com/new-method-efficiently-safeguards-sensitive-ai-training-data/