Tue, 18 Nov 25
AI Data Authentication & IP Protection Systems
- Poseidon platform for human data contribution
- Voice, art, audio data with blockchain ownership tracking
- Data pipeline validates quality and detects deepfakes
- First campaign: voice recording in multiple languages for points/rewards
- Challenge: ~10% of submissions are deepfaked voices (users training AI on own voice)
- Current deepfake detectors insufficient for nonsense detection
- Decentralized validation network approach
- Multiple validators (human + AI) vote on data authenticity
- Proof-of-stake model with rewards for correct validation, slashing for poor performance
- World ID integration for human verification (launched as mini app)
- Assumption: majority of validators are good actors
- Risk: coordinated validator attacks still being addressed
Creator Impact & AI Content Challenges
- Platform authenticity problems escalating with AI
- MetaMask seeing entire website copies including support interfaces
- Scammers using fake phone numbers in App Store reviews
- LLMs making discovery harder (search results show non-existent contact info)
- Social media verification still effective but limited
- Creator perspectives on AI tools
- Competitive advantage from AI-generated thumbnails, presentations
- Concern about platforms not disclosing AI content sources
- Art fundamentally about human connection, not just visual output
- AI as inspiration tool acceptable, unauthorized likeness replication problematic
- Need for creators to work at AI speed while maintaining human essence
Technical Primitives for Content Protection
- Key infrastructure needs identified:
- Hardware-based identity verification
- Multiplayer recognition systems for detecting inconsistencies
- Interoperability protocols across platforms to prevent IP silos
- Transparent training data attribution systems
- Economic model considerations
- Sustainable data-to-LLM-to-data feedback loop
- Creator compensation for AI training usage
- Platform responsibility vs individual artist protection
- Competition between different IP protection primitives
Chat with meeting transcript: https://notes.granola.ai/d/06930828-82c7-452c-bac9-bea0cbc25663