Copyright rulings would accelerate licensing shifts, not slow AI training
iamtrask · x · 2026-09-20
Andrew Trask and Herbie Bradley debate what copyright rulings would actually do to AI training:
- Bradley: realistic rulings wouldn't meaningfully slow training, just ramp up data spend; web-scraped pretraining is tapped out and new data is already licensed
- Trask: short-term, securing millions of licenses is a massive logistics challenge; long-term, data sources will only sell if granted more agency over the models they train — KYC, dynamic royalties, attribution
- Reality evolves from the former to the latter; rulings would only accelerate it
More from AGI Musings
- AI labs' monitoring and actual research work is now 'basically all AI', says Jeff Ladish — JeffLadish · 2026-09-20
- Luiza Jarovsky: AI won't kill all humans by 2030, but may destabilize key global systems — LuizaJarovsky · 2026-09-20
- EA critics blast 'insects matter more than people' essay: keep them away from AI control — wen_ragnarok · 2026-09-20
- Helen Toner, who voted to oust Altman, on AI safety and overseeing OpenAI — trevposts · 2026-09-20
- AI Agent Switches Auto Insurance in 5 Minutes, Saving $3,500 a Year — rohanpaul_ai · 2026-09-20
- If continual learning is nearly solved, why do we still pick reasoning effort tiers? — akbirthko · 2026-09-20