Perplexity AI Anthropic’s Settlement and the AI Industry
7 snips
Sep 15, 2025 A historic $1.5 billion copyright settlement sends shockwaves through the tech ecosystem, highlighting the debate over AI's use of data. The discussion reveals potential compliance shifts for big players and the chilling effects on startups. Legal precedents in AI and copyright are explored, revealing ramifications for authors’ rights in the evolving landscape. Additionally, challenges in image generation and source tracking are tackled, with Adobe's compensation model and a new AI app builder introduced for creators.
AI Snips
Chapters
Transcript
Episode notes
Shadow Libraries Fueled Rapid Model Gains
- Anthropic used large pirated "shadow libraries" to rapidly access millions of books for training.
- That bulk access likely improved Claude's writing tone and capabilities compared to early competitors.
Buying Books And Robot Scans
- Jaeden describes Anthropic buying many books and using a robot to scan and transcribe pages into training data.
- He frames this as a deliberate move to cover tracks and rapidly ingest large copyrighted datasets.
Fair Use Framing For Purchased Training Data
- The judge ruled that training on purchased copyrighted works can be fair use if it transforms the material.
- That legal framing separates piracy (illegal) from purchased-data training (potentially lawful).