Basically anyone who's tried to use language models for novel writing as a request rather than a completion will tell you immediately that it just doesn't pace

Basically anyone who's tried to use language models for novel writing as a request rather than a completion will tell you immediately that it just doesn't pace well at all. It basically rushes to include all the details of your synopsis/instructions in the novel as quickly as it can(extremely quickly) even if you instruct it not to do so. Not so with the old bing. The new bing though (much faster version slowly being rolled out)...fails consistently(https://imgur.com/a/vkJOhQ4 https://imgur.com/a/vkJOhQ4). With machine learning, you really can't get this drastic performance increase without switching models or increasing compute drastically. I was hoping for the best but.. For me right now, old bing is on edge dev and new bing is on edge.