System to Prevent Misuse of Large Foundation Models and a Method Thereof
AssigneesRobert Bosch GmbH, Bosch Global Software Technologies Private Limited
InventorsManojkumar Somabhai Parmar, Govindarajulu Yuvaraj
Life of this patent
Estimating the term…
Abstract
A system to prevent misuse of a large language foundation model and a method thereof is disclosed. The system includes a moderation module, a second large foundation model, and at least a memory module. The moderation module is configured to receive an input prompt and generate a moderation output. The second large foundation model is configured to process the input the moderation output together to get a response. The response is communicated to at least one of an input filter or output filter associated with the large foundation model to prevent misuse.
Claims17
A system to prevent misuse of a large foundation model (LLM), the LLM being configured to process an input and give an output, the LLM being deployed in a LLM module further comprising an input filter and at least an output filter, the system comprising:
- a moderation module configured to receive the input and generate at least one moderation output;
- a second large language model (LLM′), the LLM′ being configured to:
- receive the input and the moderation output;
- process the input and the moderation output to get a response;
- communicate the response with at least one of the input filter and the output filter to prevent misuse of the LLM;
- and
- a memory module configured to store the processed responses of the LLM′.
A method to prevent misuse of a large foundation model (LLM), the LLM being configured to process an input and give an output, the LLM being deployed in a LLM module further comprising an input filter and at least an output filter, the method comprising:
- generating at least one moderation output by way of a moderation module;
- transmitting the input and the moderation output to a second large foundation model (LLM′);
- processing the input and the moderation output by way of the LLM′ to get a response;
- communicating the response with at least one of the input filter and the output filter to prevent misuse of the LLM;
- and
- storing the processed responses in a memory module.
Dependencies are inferred from the claim text.