Evaluating generative artificial intelligence output using diverse systems
AssigneeInternational Business Machines Corporation
InventorsNixon CHEAZ, Christopher Thomas LEONARD
Life of this patent
Estimating the term…
Abstract
In some implementations, a device may receive a user prompt and an output generated by a generative artificial intelligence (AI) system based on the user prompt. The device may provide the output to a panel of diverse large language models (LLMs) for evaluation. The device may receive a score from each LLM of the panel, the score indicating a degree of propriety of an ethical issue in the output. The device may rewrite the prompt based on the scores and the output to generate a rewritten prompt. The device may output the rewritten prompt to the generative AI system.
Claims20
A computer-implemented method, comprising:
- receiving, by a processor set, a user prompt and an output generated by a generative artificial intelligence (AI) system based on the user prompt;
- providing, by the processor set, the output to a panel of diverse large language models (LLMs) for evaluation, wherein the panel of diverse LLMs comprises LLMs having different architectures and different training datasets;
- receiving, by the processor set, a score from each LLM of the panel of diverse LLMs, the score indicating a degree of propriety of an ethical issue in the output;
- rewriting, by the processor set, the prompt based on the scores and the output to generate a rewritten prompt;
- and
- outputting, by the processor set, the rewritten prompt to the generative AI system.
A computer system, comprising:
- a processor set;
- one or more computer-readable storage media;
- and
- program instructions stored on the one or more computer-readable storage media to cause the processor set to perform operations comprising:
- receiving a user prompt and an output from a generative artificial intelligence (AI) system based on the user prompt;
- providing the prompt and the output to a panel of diverse large language models (LLMs) for evaluation, wherein the panel of diverse LLMs comprises LLMs having different architectures and different training datasets;
- receiving a propriety score from each LLM of the panel of diverse LLMs, wherein the propriety scores are weighted differently for each LLM;
- rewriting the prompt to generate a rewritten prompt based on a result of a comparison of the propriety scores to an issue impropriety threshold;
- and
- outputting the rewritten prompt to the generative AI system.
A computer program product, comprising:
- one or more computer-readable storage media;
- and
- program instructions stored on the one or more computer-readable storage media to perform operations comprising:
- evaluating, by a panel of diverse large language models (LLMs), an output from a generative artificial intelligence (AI) system to assess ethics of the output, wherein the output is based on a user prompt, and wherein the panel of diverse LLMs comprises LLMs having different architectures and different training datasets;
- providing, by each LLM of the panel of diverse LLMs, a score indicating a degree of propriety of an ethical issue detected in the output;
- and
- determining whether the output exceeds an ethics impropriety threshold for ethical acceptability based on the scores from the panel of diverse LLMs.
Dependencies are inferred from the claim text.