DeepMind Researcher: Pretraining Distills Humanity's Common Good Into Model Weights

syhw · x · 2026-11-20

DeepMind researcher ibab proposes an ethics framing for LLMs that deserves more attention: pretraining distills vast amounts of humanity's collective work from the internet into model weights. That content was created before LLMs were conceivable and released freely as a common good for others to read — raising the question of whether permission to read extends to permission to train. syhw strongly agrees.

Related event: Google Researcher Argues LLM Weights Should Be Open-Sourced as Repayment to the Commons(3 posts)→

Original post →

More from AGI Musings

AGI Musings channel →