AI 生产的低质内容多过价值

gary marcus cites ft and washington post data showing ai has flooded markets with books, apps, and music, yet sales and quality have declined while gdp remains stagnant. this discrepancy suggests ai functions as a word predictor producing 'slop' rather than value, with leading labs burning billions on token costs despite minimal roi. mathematicians have raised similar concerns via the leiden declaration, fearing ai-generated proofs will overwhelm verification systems. marcus concludes that despite coding being a potential exception, the ai sector is largely an economic bubble where vendors lose money on every transaction.