Automatic Construction of a Korean Toxic Instruction Dataset for Ethical Tuning of Large Language Models
This addresses the need for ethical AI in Korean-language applications, though it is incremental as it focuses on dataset creation rather than a new method.
The paper tackles the problem of mitigating unethical language generation in Large Language Models by automatically constructing KoTox, a dataset of 39K toxic instruction-output pairs, to improve ethical tuning and promote safer interactions in NLP.
Caution: this paper may include material that could be offensive or distressing. The advent of Large Language Models (LLMs) necessitates the development of training approaches that mitigate the generation of unethical language and aptly manage toxic user queries. Given the challenges related to human labor and the scarcity of data, we present KoTox, comprising 39K unethical instruction-output pairs. This collection of automatically generated toxic instructions refines the training of LLMs and establishes a foundational framework for improving LLMs' ethical awareness and response to various toxic inputs, promoting more secure and responsible interactions in Natural Language Processing (NLP) applications.