Opening Keynote Day 3: Taking Decentralized AI Mainstream
summary
Vana is a live platform that turns user exports into permissioned, tokenized datasets called data dials and verifies inputs with proof of contribution running in TEEs; examples include a 140,000-user Reddit dial, a Tesla car dial (DLP Labs), and a Spotify proposal that sold data to Solo AI. Contributors earn VRC-20 data tokens (early Spotify contributors earned ≈$1,000; high‑karma Reddit users ≈$300–$400). Federated pretraining via Flower (Collective One) aims at multi‑platform models using pooled exports; secure compute rents model jobs instead of handing over raw data.
AI-written summary · about
chapters (9)
- 0:00Decentralized AI introduction
- 3:19AI training data wall
- 7:53Self-sovereign AI and data
- 10:01What is a data dial
- 16:21Data tokens and finance
- 19:49Training AI with pooled data
- 25:31Data portability and sovereignty
- 28:42Data serfdom and AI economics
- 31:33Monetizing data & Q&A
related talks
33mKeynote: Fine-tuning our way towards openness in AI
40mOpening Keynote: AI & Platform Teams
04mCES Interview
32mProgress In the Decentralized AI Stack
35mBeyond Centralized Models
13mEmpowering Developers through Open-Source AI
01h 16mLifting the Hood on Big AI - The Future of Transparency and Accountability in AI
34mKeynote Panel Discussion: Signal vs. Noise: Betting on AI’s Next Wave