by Brian Wang from NextBigFuture.com on (#6QKX1)
Facebook has ~10x the proprietary language data on database as was used to train the LLaMa models. In images Facebook have 20x more than that. Instagram and Youtube have 2x more that in uploaded video. Tesla's data capture-ability dwarfs all (at 20x more again) than Youtube. Size matters Facebook has ~10x the proprietary language data ... Read more