Technological and Institutional Hindrances to the Dissemination of Large Language Models

Abstract

With the advent of generative AI, machine learning technologies play an integral role in the digitization of society. Large language models (LLMs), a representative of the generative AI, are not free from issues regarding to social and ethical viewpoints, such as hallucination, or violation of privacy rights or copyrights. These problems originate from the basic mechanisms of LLM and are difficult to eliminate completely. Thus, engineering efforts are mandatory to complement the deficiencies of LLM, and a technology supply chain involving diverse actors is taking shape. In view of legal aspects, European AI-ACT is supposed to have a huge impact on the LLM business. However, the legal liability for the proliferation of technology leading to misuse or compensations for damages still remains uncertain. AI governance that all the actors, along the technology supply chain, work collaboratively on quality management of Trustworthy AI will reduce the hindrances to the dissemination of LLM.