Questions

• Does Polotno custom metadata support MobX reactivity?
  • Potential for intelligent incremental re-generation suggestions
    • E.g. you just edited the topic, “We auto-magically generated X, Y, and Z for you using the previous topic. Would you like us to re-generate them with the new topic?”

Considerations

• Might make sense to bring forward the state system for incremental regeneration
  • Person emoji → >75% human text
  • Technologist emoji (person w/ laptop) → 25%-75%
  • Laptop emoji → >75% AI text
  • Warning emoji (dismissible?) → Dependencies haven’t been re-generated
    • Of course, warn of any non-AI (human-edited) responses that will get overwritten after re-generation
  • […?] emoji → out of date, AI-generated and an ancestor was edited
  • Might take too long though…

Editor difficulty levels

• Can probably combine super easy editor stage with the import process
  • Then if they want more control, they open the editor from the import interface
• Basic editor is just Polotno, and if you open the special custom options you can edit the metadata as needed
• Advanced flow-chart editor will probably take too long, not be useful enough

Goal

• Given document text (and images?), get a concise topic description

• E.g. give it a bunch of text about WWII with a focus on specific battles, it might return “World War II, with a focus on specific battles”

• Bag of words

• Just feed all the text to an LLM, ask it to describe the topic in only a few words

  • Call it a day

Thoughts

• I wonder if there’s any accuracy in OpenAI’s BPE tokenizers, using those numerals to correlate to readability