Abstract
— The adoption of large language model-based applications has been recognized to be future of the adoption of artificial intelligence. However, large language model (LLM) based applications benefit from an uncompensated volunteer input. The uncompensated volunteer input arises from the core service tasks such as output preference indication from a large number of subscribers. Given that the large-scale adoption of LLMs is poised to trigger job loss, the challenge arising from the uncompensated volunteer input should be addressed. This is especially important for the case of developing countries. The presented research proposes a network architecture with cost consensus capacity in this regard. In the presented research, cost consensus capacity refers to the consideration of the LLM query heterogeneity and varying usefulness in the future for other LLM system subscribers in determining the cost associated with the volunteer's work. The proposition of the network architecture recognizes, addresses and ensures an avoidance of an artificial intelligence winter. In this case, a cost reduction is applied to the case of compensation of volunteer work input. In the proposed approach, the compensation is not given to the volunteer but the sovereignty authority in the volunteer location. The application of a cost reduction factor to prevent artificial intelligence winter reduces costs by an average of (45.5-78.2)%.