Meta Accused of Illegally Downloading Over 80TB of Books from LibGen, Anna's Archive, and Z-Library for AI Training
Meta has been exposed for illegally downloading over 80TB of books from LibGen, Anna's Archive, and Z-Library to train AI models.
In comparison, in 2010, Aaron Swartz downloaded 70GB of articles from JSTOR (only 0.0875% of Meta's download volume) and faced a $1 million fine and 35 years in prison, ultimately committing suicide in 2013.
This comparison has sparked widespread discussion about the data acquisition methods of tech giants and the legal consequences.
Source: Public Information
ABAB AI Insight
Meta previously trained its Llama series models through large-scale data collection, and this exposure continues its aggressive strategy of acquiring massive data for AI training, similar to controversies faced by companies like OpenAI regarding the use of pirated content for training.
In terms of capital pathways, Meta reduces costs by obtaining training data through non-public channels, with funding skewed towards expanding model parameter scales, while individual actions during Swartz's time faced severe penalties, highlighting differences in enforcement standards.
Similar to recent controversies over AI training data sources, this incident occurs during a phase of restructuring copyright and data acquisition rules in the AI era.
Essentially, it represents a reconstruction of data capital: tech giants accelerate AI development by acquiring massive training data through gray channels, concentrating pricing power in platforms with strong data acquisition capabilities, while traditional copyright laws struggle to keep pace with the scaled demands of AI, resulting in severe penalties for individuals and small-scale actions while allowing greater operational space for giants.
ABAB News · Cognitive Law
The larger the data scale, the stronger the rule elasticity.
Individuals go to jail for downloads, while giants download for training.
In the AI era, data is oil, and acquisition is king.