Extending GPT-3’s context window infinitely
erwinmayer.medium.comWhat do you think of this idea to infinitely extend GPT-3's context window, by storing context in GPT-3 itself or in secondary layers?
I think GPT-3 is the most overrated development in the 2020s.
It's great at getting answers that are almost right and it reveals dangerous vulnerabilities in our "social software".
That is, it can do a great job of creating shitposts like Elon Musk's tweets or opinion columns for the New York Times. It can simulate the writings of a high-status person who is never held accountable for being right or wrong.
It can't do the job of a proofreader, janitor or anybody else who is judged on doing their job right.