Tag
前Meta研究员Stephen Roller在法庭证词中承认在职期间通过BitTorrent下载并分发了盗版Books3数据集，该数据集被用于训练Llama模型，尽管Meta此前否认通过BitTorrent获取该数据集。Meta回应称该下载可能是为个人用途，而原告计划要求法院进一步查明Meta获取和使用该数据集的详情。
MIT Technology Review 的每日科技简报，涵盖 AI 实验室收购倒闭初创公司数据用于训练、AI 助力的跨物种交流研究，以及月球基地建设的挑战。
Data brokers are buying datasets from SaaS companies for $100,000 to $2 million, then reselling them to AI training companies and frontier labs to integrate SaaS capabilities into AI applications, while SaaS companies use the proceeds to fill revenue gaps.
科学家们日益担忧AI公司可能在未经允许的情况下将用户上传的研究稿件、想法和数据用于模型训练，引发关于研究隐私和知识产权归属的讨论。相关争议包括AI公司被指控抓取用户上传文档，以及OpenAI等公司展示模型研究成果时引发的伦理质疑。
帖子讨论AI实验室面临合成数据过剩、却极度缺乏真实世界失败日志数据的困境，指出能捕获并利用失败数据的玩家将掌控AI评估层，形成真正的护城河。
The author comments on the perceived double standard in the AI copyright debate, arguing that while humans learn from existing works to create, similar behavior by machines is treated as unprecedented IP infringement.
A thread arguing that AI training data is a uniquely perishable product that loses value once models master a task at around 70% accuracy, fueling a hot market that minted young billionaires even as the largest labs bring data creation in-house.
USA Today and several of its local newspapers sued OpenAI, seeking over $250 million in damages, alleging the company copied hundreds of thousands of articles without permission to train its AI models.
OpenAI reportedly plans to release hundreds of AI-solved math problems, drawing criticism from mathematicians who argue such outputs devalue human mathematical reasoning and proper proof standards.
A commentary noting Ben Affleck's claim that he built a private eight-month filmed dataset to keep filmmakers' ownership, arguing that most open AI models were still trained on peers' work without consent — a gap that undercuts smaller studios.
Oxford University allowed OpenAI to use texts from the Bodleian Library, including old PhD theses, to train its AI models, raising concerns about ethics and energy use despite the library's aim to make texts more accessible.
Japanese used bookstores are seeing a 5x sales surge due to bulk purchases suspected to be for AI scanning and destruction by foreign AI tech giants, raising concerns over data sourcing ethics.
Snorkel AI has tripled its valuation to $3.5 billion after raising $350 million in Series E funding, driven by surging demand for AI training data. The company now focuses on data-as-a-service, providing complete datasets and simulated environments through synthetic and expert-generated data.
Garry Tan tweets that Facebook's Muse is generating valuable AI training data from user behavior, comparing its data collection to the Onavo app.
The tweet highlights Facebook's potential to collect vast amounts of personal AI training data through its 'muse' product, comparing it to Onavo and suggesting it could significantly boost their data flywheel.
The article explores why AI models from different labs have distinct strengths, like coding or writing, despite potentially using training data from the same vendors. It notes that US data vendors also supply Chinese labs, highlighting concerns about overlapping training resources.
AfterQuery, an AI training-data startup, has become Y Combinator's fastest unicorn with a $3.2 billion valuation, just 18 months after launching. The company works with major AI labs and employs knowledge professionals for model training.
AI companies are secretly buying and destroying physical books to obtain training data, locking knowledge away, and Anna's Archive is urging volunteers to scan rare books to prevent cultural heritage loss.
Google is acquiring employee data from bankrupt Spirit Airlines, raising privacy concerns from flight attendants who fear their data may not be sufficiently protected.
Gregory Kurtzer, founder of Rocky Linux, announces OpenWALDO, a project aimed at opening up AI training data to foster true open-source AI development.