Meta's Llama 3 Trained on Pirated Data: Internal Documents Reveal Zuckerberg's Approval

Newly unsealed documents reveal that Meta trained its Llama 3 large language model using copyrighted material from the pirated library Library Genesis (LibGen). Despite internal concerns, CEO Mark Zuckerberg approved the use of this data. This decision exposes Meta to potential copyright lawsuits and negative publicity, highlighting broader concerns about the ethical sourcing of data in AI development.