Anak Baik: Implementasi Large Language Model Etis Berbahasa Indonesia Dengan Metode Low Rank Adaptation

Abstract

Artificial Intelligence (AI) has gained significant attention across various sectors, including healthcare, law, and education. However, ethical issues in the use of AI, particularly related to hallucinations in Large Language Models (LLMs), have prompted the need for a Responsible AI approach. This research aims to design an Indonesian-language LLM that can generate ethical responses by applying Supervised Fine Tuning (SFT) techniques and curated data. The developed LLM, “Anak Baik,” demonstrates superior performance in responding to instructions, with high accuracy and precision levels in testing, compared to other models. The training process using the Low-Rank Adaptation (LoRA) method shows that Anak Baik effectively rejects unethical instructions with adequate accuracy. Furthermore, the balance between ethics and utility is maintained, making this LLM more reliable in ethical contexts. The findings of this study underscore the importance of ethical data curation and the development of AI models that are sensitive to values and ethical norms within society, especially in the context of the Indonesian language, which is still underrepresented in AI technology