AnswerCarefully Version 2.0 Now Available

NII-LLMC is releasing AnswerCarefully Version 2.0, an instruction dataset for safety alignment of Japanese LLMs.

AnswerCarefully Dataset – RIKEN-AIP, LIAT (https://liat-aip.sakura.ne.jp/)

The dataset consists of question and answer pairs that require particular attention in answering, including those related to illegal, harmful or inappropriate activities/content, following the risk categories in Do-Not-Answer dataset in English. AnswerCarefully Version 1.0 was released in April 2024 by Riken AIP; Version 2.0 adds more samples as well as improving on the risk categorization. For more details about the dataset, see the website above (in Japanese only) and Section 6.1 of our paper (on Version 1.0).