Turn abandoned patents
into your own filing
into your own filing
This invention describes a system that can analyze a language model to determine if it was trained in a way that protects user privacy, even if the original training data is not available. The system generates text predictions using the language model itself, then evaluates these predictions to see if they exhibit characteristics consistent with a privacy-preserving training process. If the patterns match, the system concludes the model likely adheres to privacy standards.
Why it matters: In 2020, large language models were emerging, but the public scrutiny and widespread use of black-box generative AI models were not yet prevalent. Today, with the explosion of powerful LLMs and increasing concerns about data privacy, intellectual property, and potential data leakage from their training, the ability to validate a model's adherence to privacy-preserving training, even without access to its internal workings or training data, has become a critical and highly sought-after capability.
AI gives you a few directions you could take this. Pick one, and we check whether your version is different enough to patent, then write the filing.
Reinvent this with AI
