Raw LLM Responses

Hello, nice work. One limitation is the lack of GPT-4 originality and it's difficulty to hold it when suggested. I have made a try with imposing each time a specific genre or a trope or a character with each time quite loose freedom of choice and it just gave me quite similar outline, objects or villain, just with other settings or genre. For exemple a old scroll appear in each of its answers. If we give it a more full personal outline, I find it loose the ability to properly deal with it and I ended to create all the stuffs quite manually (e.g. The Opening and Final paragraphs) because it looks to want all the time keep on track with its training data. Gpt-3.5 was less stubborn but lack of token memory and outputs. Then, one problem for people wanting to automate something like this with the current available ai will be that only the first story generated will be quite 'original' all new will be some kind of plagiarized works. Even phrasing outputs are at some point repetitive in their structure, choosing à voice to rewrite doesn't really fix it, it just hide a bit better the lack of uniqueness