I know, I know. But I don't say this lightly. I've worked in NLP/G for nearly 15 years and after spending the last 6 months working with GPT-3 I feel the writing is on the wall.
But it isn’t. I am still actively involved in NLG both professionally and academically and neural NLG systems have great promise but are still far from actively delivering tangible solutions to areas such data-to-text NLG. Inaccuracy, hallucinations are still highly problematic.