Are large datasets too much exposure or worth the risk? For OpenAI it appears the latter.