Core Viewpoint - The ongoing legal disputes surrounding OpenAI highlight the critical issue of copyright infringement in the context of generative AI, particularly regarding the use of copyrighted materials for training models [1][2]. Group 1: Legal Cases - OpenAI faces two significant lawsuits: one in Germany for copyright infringement involving music lyrics, and another from The New York Times regarding user logs [1][2]. - The German court ruled that OpenAI's use of copyrighted lyrics to train ChatGPT constitutes infringement, marking a pivotal moment in AI copyright law [1][3]. - The New York Times lawsuit demands OpenAI to submit up to 20 million user logs, which could provide crucial evidence in determining copyright infringement [2][6]. Group 2: Legal Precedents - The German court's decision challenges the previous notion of "fair use" in AI training, asserting that the model's "memory" and output of lyrics are forms of reproduction [2][5]. - The ruling emphasizes that the training process of AI models is not as opaque as previously thought, suggesting that the technical details can be scrutinized under copyright law [2][5]. - The court's interpretation aligns with EU copyright directives, indicating that any form of reproduction, even if indirect, falls under copyright infringement [5][8]. Group 3: Implications for AI Companies - The legal outcomes suggest a shift towards stricter compliance for AI companies, necessitating prior authorization for using copyrighted materials in training [8][9]. - OpenAI's defense strategy, which claims that the model does not store or replicate content but learns from it, has been rejected by the German court [4][5]. - The ongoing legal battles may set a precedent for how AI companies operate, potentially ending the era of "free lunch" in AI training without proper licensing [8][9].

编辑丨肖潇 大模型"偷书"算不算"偷"？是贯穿生成式AI发展过程中的议题，近几年版权纠纷的熊熊战火足以说明其重要性。 本月，OpenAI分别在欧洲大陆和美国本土迎来两起诉讼，把这个问题又推向了新的高度： 一起是被称为欧洲首例生成式AI版权侵权案，据央视报道， 当地时间11月11日，德国慕尼黑第一地区法院裁定：OpenAI未经许可使用受著作 权保护的音乐歌词训练ChatGPT模型，已构成侵权。 记者丨王俊 图/德国慕尼黑法院就GEMA v. OpenAI案的判决书 另外一起则是已经缠斗两年的《纽约时报》诉OpenAI案，11月7日，纽约南区联邦法院法官Ona T. Wang发布最新命令，责令OpenAI向《纽 约时报》提交多达2000万条ChatGPT用户日志。 业内对这两起案件倾注了足够的关注，仔细翻阅法律文书，有很多高光点不容忽视： 1、推翻"合理使用" 此前的不少判决中，大模型训练被视为"合理使用"，即在特定情况下，模型无需得到授权即可使用作品。但德国法院率先认定，AI模型"记 忆"与"输出"歌词均构成复制侵权，将AI"学习"定性为"偷书"。 技术祛魅，大模型训练数据的过程并不神秘 生成式AI的训练过程是 ...

Core Viewpoint - The ongoing legal disputes surrounding copyright infringement related to generative AI, particularly OpenAI's ChatGPT, highlight the complexities of copyright law in the context of AI training and usage, with recent rulings indicating a shift in how AI companies may need to operate regarding copyright compliance [1][10]. Group 1: Legal Cases Overview - OpenAI faces two significant lawsuits: one in Germany regarding unauthorized use of copyrighted music lyrics for training ChatGPT, and another from The New York Times concerning user logs [1][2]. - The German court ruled that OpenAI's use of copyrighted lyrics constitutes infringement, challenging the previous notion of "fair use" in AI training [2][6]. - The New York court has ordered OpenAI to submit up to 20 million user logs, emphasizing the importance of evidence in copyright cases [7][8]. Group 2: Technical and Legal Implications - The German court's decision indicates that AI's "memory" of lyrics is equivalent to copying, as the model retains fixed parameters that can reproduce copyrighted content [4][6]. - The legal scrutiny is increasing, with courts demanding transparency in AI training processes, which may limit AI companies' defenses based on technical complexity [3][5]. - The ruling suggests that AI companies must adhere to copyright laws, marking a potential end to the "free lunch" era for AI, where training could occur without prior authorization [9][10]. Group 3: Industry Reactions and Future Directions - Legal experts indicate that the requirement for extensive evidence submission in copyright cases is becoming more common, particularly in AI-related disputes [8]. - OpenAI's stance on user privacy and the protection of user data is being challenged by the demands of the lawsuits, which could set a precedent for future cases [9]. - The outcomes of these cases may redefine the operational landscape for AI companies, necessitating a shift towards obtaining licenses before training models [10].