AI開発OpenAI Blog8月7日From hard refusals to safe-completions: toward output-centric safety training元の記事を開く要約を生成中です...メモを読み込み中...