Wed Apr 20 12:14:56 2022
(*cfaf6bd4*):: our phones can smelll us. The easterners have to smell us and it is overwhelming
(*cfaf6bd4*):: https://atxcf.slack.com/archives/C01AH8CV0TE/p1650456802237769https://atxcf.slack.com/archives/C01AH8CV0TE/p1650456802237769
*** Using writer & [[AI]] for an [[Open Source Musical]]
+public!
(can’t stop looking into multi-story management, fml)
# TaleBrush
As the writer generates multiple stories, TaleBrush stores each. Users can compare and choose among them (Figure 5). A dropdown menu lists the generated stories, which will be displayed on the canvas with low opacity. As the writer moves the mouse over the list, the hovered story is shown in the text box and highlighted on the canvas. The writer can select one of these to ‘roll back’ to a past generation.
*we focus on visual encodings that can both express high-level changes in the story’s progression but also be manipulable.*
TaleBrush can potentially be extended to support the authoring of high-fidelity, longer text. To provide such support, the tool would need to consider writers in the translation stage, where ideas and plans are transformed into detailed text.
• generative tools should be able to follow the writer’s specifications. The annotation approach would also need to be reconsidered, such as annotating the fortune of the character after summarizing the long text with crowdsourcing or algorithms. Moreover, the control and sensemaking should be re-designed for longer text. For example, a tool might allow the writer to first sketch an outline and then a detailed story. A more complete tool might enable a long-term study of how co-creation impacts writing.
Visualizing and Visually Expressing Stories
• To produce content, some tools connect the author to other humans or support collective story writing by writing stories with the crowd. These systems can structure creative leadership [77] and decision-making [78], or even simulate the characters with crowds [63]. Newer tools have adopted novel machine learning (ML) and natural language processing (NLP) techniques [153]. Among those techniques, ML’s generative functions offer a powerful alternative approach. These systems append or suggest generated texts the writer’s text using sophisticated language models [24, 28, 29, 129, 143]. In theory, these algorithms can act as writing support tools. However, limited controls and interactions—largely rephrasing prompts and contexts—constrain their applicability. Our goal with TaleBrush is to leverage these language models but provide an alternative control strategy using both text and abstract visual representations and interactions.
To realize this interaction, we implemented a technical architecture that achieves steerable story generation ( **which i’m working on in Roam Research (preparing to export to a Github Repo).
Machine Story Generation
• Reflections on generalizable design elements of control interaction for iterative human-AI co-creation and how line sketching interaction can be expanded to other generative contexts.
• *A GPT based controllable language model architecture that generates story sentences with the sketching input on the protagonist’s fortune.*
- U014EH1SZKL
- bitcoin-the-musical
Using writer & [[AI]] for an [[Open Source Musical]]
+public!
(can’t stop looking into multi-story management, fml)
# TaleBrush
https://johnr0.github.io/assets/publications/CHI2022-TaleBrush.pdf
As the writer generates multiple stories, TaleBrush stores each. Users can compare and choose among them (Figure 5). A dropdown menu lists the generated stories, which will be displayed on the canvas with low opacity. As the writer moves the mouse over the list, the hovered story is shown in the text box and highlighted on the canvas. The writer can select one of these to ‘roll back’ to a past generation.
we focus on visual encodings that can both express high-level changes in the story’s progression but also be manipulable.
TaleBrush can potentially be extended to support the authoring of high-fidelity, longer text. To provide such support, the tool would need to consider writers in the translation stage, where ideas and plans are transformed into detailed text.
- generative tools should be able to follow the writer’s specifications. The annotation approach would also need to be reconsidered, such as annotating the fortune of the character after summarizing the long text with crowdsourcing or algorithms. Moreover, the control and sensemaking should be re-designed for longer text. For example, a tool might allow the writer to first sketch an outline and then a detailed story. A more complete tool might enable a long-term study of how co-creation impacts writing.
Visualizing and Visually Expressing Stories
- To produce content, some tools connect the author to other humans or support collective story writing by writing stories with the crowd. These systems can structure creative leadership [77] and decision-making [78], or even simulate the characters with crowds [63]. Newer tools have adopted novel machine learning (ML) and natural language processing (NLP) techniques [153]. Among those techniques, ML’s generative functions offer a powerful alternative approach. These systems append or suggest generated texts the writer’s text using sophisticated language models [24, 28, 29, 129, 143]. In theory, these algorithms can act as writing support tools. However, limited controls and interactions—largely rephrasing prompts and contexts—constrain their applicability. Our goal with TaleBrush is to leverage these language models but provide an alternative control strategy using both text and abstract visual representations and interactions.
To realize this interaction, we implemented a technical architecture that achieves steerable story generation ( **which i’m working on in Roam Research (preparing to export to a Github Repo).
Machine Story Generation
- Reflections on generalizable design elements of control interaction for iterative human-AI co-creation and how line sketching interaction can be expanded to other generative contexts.
- A GPT based controllable language model architecture that generates story sentences with the sketching input on the protagonist’s fortune.
(*cfaf6bd4*):: I would be impressed with DALLE if i could just get it to give me a handjob. But it can’t, thus shitty software