[2411.01271] Interacting Large Language Model Agents. Interpretable Models and Social Learning

View a PDF file from the paper entitled The Large Language Model Factors. Interpretation and social learning models, by Jin and other authors

PDF HTML (experimental) view

a summary:This theoretical paper and algorithms are discussed for the interaction of LLMAS model factors using methods of processing statistical signals and partial economy. While both fields are mature, their application to decision -making that involves the LLMAS reaction is still not explored. Motivated by analyzing the bizi feelings on the online platforms, we build interpretable models and algorithms that enable LLMAS to interact and perform the Bayezi reasoning. Since the LLMAS reaction learns from both previous decisions and external inputs, they can show bias and grazing behavior. Consequently, the development of interpretable models and random control algorithms is necessary to understand and mitigate these behaviors. This paper has three main results. First, we appear using Bayesian preferences to disclose partial economy that individual LLMA meets the necessary conditions and sufficiently to glorify the rational (rationalism), and given the observation, LLMA chooses a procedure that increases organized benefit. Second, we use Bayesian to build interpretable models for LLMAS, which interacts successively with each other and the environment during the performance of the Baysi reasoning. Our suggested models pick up the grazing behavior shown by LLMAS interaction. Third, we suggest a framework for random control to delay grazing and improve the accuracy of the state under two: We explain the effectiveness of our methods on real data collections to classify hate speech and evaluate the quality of the product, using open source models such as Lama and closed source models such as ChatGPT. The main meals of this paper, which are based on experimental and mathematical analysis, are that LLMAS acts as the same rationality, which shows social learning when interacting.