[Question] How does OpenAI’s language model affect our AI timeline estimates?

OpenAI re­cently an­nounced progress in NLP, us­ing a large trans­former-based lan­guage model to tackle a va­ri­ety of tasks and break­ing perfor­mance records in many of them. It also gen­er­ates syn­thetic short sto­ries, which are sur­pris­ingly good.

How sur­pris­ing are these re­sults, given past mod­els of how difficult lan­guage learn­ing was and how far AI had pro­gressed? Should we be sig­nifi­cantly up­dat­ing our es­ti­mates of AI timelines?