Matthew Garrett on Nostr: ML models being trained on information they can scrape from the internet provides a ...

ML models being trained on information they can scrape from the internet provides a perverse incentive - if knowledge is shared in ways that can't easily be scraped, that knowledge remains outside the scope of what the models can learn from. People working in fields that are currently poorly documented suddenly have a reason to stop writing blog posts, and instead engage in spoken presentations. A new kind of passing down of secrets from generation to generation.