Self-correcting generative software for prompt processing

Abstract

Large language models have difficulty understanding the context, have discrepancies with the task at hand, and quite often make mistakes. A person is forced to interfere quite often in AI work, which is ineffective for large and time- consuming tasks. This study suggests the following: AI teams and coordination of their work, which interact with each other to achieve a set goal. In this study, this will be achieved by passing messages between AI through plain text. The implementation includes using the API of a popular messenger to wrap the AI into a certain bot, which will be assigned certain roles. Each such bot will be assigned its roles and requests, and they, guided by their role, request, and task, will jointly solve the task assigned to the group. The expected results are a significant reduction in human intervention, increased automation, system performance and reliability, and a wider range of applications.