近日，知名媒体TechSpot报道，Meta公司承认在使用“盗版”书籍来训练人工智能（AI）的过程中，并未支付相应费用。在一桩诉讼中，Meta坦承利用了知名开源图书数据集Books3，以及众多其他材料，以训练其Llama 1和Llama 2大模型。
Books3是一个包含近20万本书的纯文本集合，总容量近37GB的大型数据集。然而，Meta方面表示，其在未经授权的情况下使用受版权保护的作品训练大模型，并不需要原作者的同意、许可或付费。他们主张，任何未经授权复制Books3中受版权保护的作品都应被视为“合理使用”。
这一立场引发了业界和原作者的广泛关注。一方面，Meta的行为被认为是对版权的侵犯，另一方面，其将版权作品用于训练AI是否属于“合理使用”，尚无明确法律规定。这也使得Meta在这一争议中处于风口浪尖。
针对这一事件，我国相关部门表示，将密切关注Meta的行为，并敦促其尊重知识产权，合法使用版权作品。同时，也将进一步完善相关法律法规，为AI技术的发展提供有序的法律环境。
英文翻译：
News Title: Meta Admits to Using Pirated Books to Train AI, Sparking Controversy
Keywords: Meta, Pirated Books, Artificial Intelligence, Controversy
News Content:
Recently, TechSpot reported that Meta admitted to using “pirated” books to train artificial intelligence (AI) without paying the corresponding fees. In a lawsuit, Meta confessed to using the well-known open-source book dataset Books3, as well as many other materials, to train its Llama 1 and Llama 2 large models.
Books3 is a large-scale dataset containing nearly 200,000 books in pure text, with a total capacity of nearly 37GB. However, Meta argues that the use of copyrighted works for training large models does not require the original author’s consent, licensing, or payment. They claim that any unauthorized copying of copyrighted works in Books3 should be regarded as “fair use.”
This position has attracted widespread attention from industry professionals and original authors. On the one hand, Meta’s behavior is considered copyright infringement, and on the other hand, whether the use of copyrighted works for AI training falls under “fair use” remains unclear under existing laws. This controversy has placed Meta in a delicate position.
In response to this event, relevant domestic departments have stated that they will closely monitor Meta’s actions and urge it to respect intellectual property rights and use copyrighted works legally. At the same time, they will also further improve relevant laws and regulations to provide an orderly legal environment for the development of AI technology.
【来源】https://www.techspot.com/news/101507-meta-admits-using-pirated-books-train-ai-but.html
Views: 1