The exponential benefits of using transformer models for natural language generation have led to a lack of focus on algorithm design and system building, and more emphasis on acquiring vast amounts of data to improve models. The emphasis is now on the cost of development rather than the search for new breakthroughs.