A method for preparing evaluation models of a computer device comprising at least one processor, comprising:
- registering artificial intelligence models for evaluation to be used by characteristics of evaluation target data in an evaluation system, by the at least one processor;
- confirming at least one of uniqueness and parameter information of the artificial intelligence models for evaluation, by the at least one processor;
- setting prompts for evaluation to be input into the artificial intelligence models for evaluation, by the at least one processor;
- determining an output system of output values of the artificial intelligence models for evaluation, by the at least one processor;
- connecting input pipelines for input of the artificial intelligence models for evaluation, by the at least one processor;
- connecting output pipelines for process of output values of the artificial intelligence models for evaluation, by the at least one processor,
- wherein the evaluation system is configured to
- receive evaluation target data and preprocessing it by characteristics of each of a plurality of artificial intelligence models for evaluation;
- control input of the preprocessed evaluation target data into the plurality of artificial intelligence models for evaluation;
- configure each of the plurality of artificial intelligence models for evaluation into a plurality of instance models in multi-dimension by characteristics of the evaluation target data, each of the plurality of instance models corresponding to a specific classification within at least one of the characteristics of the evaluation target data;
- and
- integrate evaluation results for the preprocessed evaluation target data of each of the plurality of artificial intelligence models for evaluation and generating an overall evaluation result,
- wherein the evaluation system is configured to control the input by providing the same preprocessed evaluation target data as a common input to each of the plurality of instance models such that each of the instance models evaluates the common input from a different perspective associated with its corresponding specific classification,
- wherein the characteristics of the evaluation target data comprise at least two of age, nationality, player role, language, culture, inclusion of visual data, gender, and hearing impairment,
- wherein the evaluation system is configured to generate the plurality of the instance models by generating at least two instance models for each of the artificial intelligence models for evaluation in association with each of the least two characteristics based on classification criterion predetermined for each of the least two characteristics,
- wherein each of the plurality of instance models outputs scores or review comments as an evaluation result for the preprocessed evaluation target data, and the evaluation result is stored in a form of ledgers of blockchain,
- wherein the evaluation system is configured to
- generate the prompts for evaluation to be input into each of the plurality of instance models based on evaluation-related policies that manage sensitive content including at least one of race, region, or politics,
- conduct additional evaluation for the evaluation target data using a large language model (LLM) based on the evaluation result of each of the plurality of instance models, input to the LLM, and
- generate the overall evaluation result based on the additional evaluation,
- wherein the LLM functions as a supervisor model that receives the scores and the review comments from each of the plurality of instance models as a plurality of evaluation opinions, and conducts the additional evaluation by generating a supervisory evaluation opinion based on said plurality of evaluation opinions, and
- wherein the overall evaluation result comprises scores for the additional evaluation and review comments for the additional evaluation.