distillation-stealing

Is distillation really NOT stealing?

The typical argument goes like model makers train on scraped data and if we consider that appropriate then surely distillation, which is basically scraping model answers and learning from them is also okay. Complaining about distillation by large model makers can thus be deemed hypocritical… Or can it?

Recently Google lost a lawsuit to a company scraping their search results. The judge deemed search results non copyrightable.

But let’s think about other publicly available data that could technically be collected, organized & processed by anyone but practically won’t be and such a person would rather pay for the end result than do that themselves AND where gaining access to that end product would clearly constitute stealing

Bloomberg Terminal credit scoring db Maps