The San Francisco-based artificial intelligence lab scraped content from more than 10 million articles, nearly a third of which came from the New York Times alone.