Yesterday I read through some of the most excellent gradio documentation where @freddyaboulton had mentioned iterative outputs and how to create UI in gradio to do synchronous and asynchronous event based interactions.
Then after reading the article and docs and wanting to learn more about the live feature in gradio, I played with the calculator example, some generator models for text (finally realizing how to use access tokens properly), then wrote a simple program that persists generated mindfulness stories with output to datasets called StoryWriter:floppy_disk:.
I found with examples it is fairly good at generating interesting exportable / editable stories using a mix of operators and prompting using prepared examples for prompt intros and outros.
I had some HF questions:
If I wanted to personalize my save for a user addressing the pain of keeping their chosen results clean for their own views, is there a way to read and save a “user name” or possibly a cookie user driven account name that would allow filtering of output in dataset so users can view just their own stuff?
Given state of transformers and state of art for story generation what are the communities thoughts on tips for a narrative editor AI pipeline strategy that might be able to write/rewrite based on sentiment and emotional intelligence of what the user likes and is going for in their generative story?
Useful articles and papers on the subject are here:
There are a number of successful format targeting AI programs on open market that are inspiring in capability like copy.ai, simplified.com, Jasper, Deepstory, Rytr and Inferkit.
- With success of similar human augmented AI pipelines on media outlets too like the Dudesy podcast - I’m interested if transformers has ever been extended to use sentiment in NLP text attention and generation similar to how writers can use wheel of emotions to write better fictional narratives like here: How to Use the Wheel of Emotions to Write Better Fictional Characters | by Jason Tweed | The Startup | Medium ?