Should add a cheetsheet for AI, and idea of dynamatically generate UI with wikitext

[linonetwo]

Tiddlywiki is good at filtering and transcluding, so why not create an all-in-one tiddler that has all necessary info for AI to read? And gradually remove parts that is well-known by AI.

@linonetwo

Next 10 years we will observe AI’s exponential growth, PR with AI, review with AI, users expect we should have AI integration, and even AI makes manually taking note not longer necessary.

I think in the coming years, predefined UI will gradually become unnecessary, except for very complex UI like calendar, WYSIWYG editor, whiteboard. AI will help assemble user interface out of void dynamatically . So a simple core is necessary.

For example, Do we even need a $:/ControlPanel UI predefined? Maybe even not necessary, let AI modify the config tiddlers, or write a UI out of nothing when we need it.

From my perspect of view, TiddlyWiki is a graph database. And plugins provides some predefined UI to do CRUD with it. I’m still writing UI with AI’s help, so normal users won’t need to, but in the next few years, normal users will be able to DIY, and discard the UI when it finish the job, because AI token will be so cheap (like what DeepSeek does), discard an AI result will feel like Exhale after Breathe. (That is why I don’t think AI result should be store in the wiki persistently.)

I’m using Claude Opus4.5 which could write wikitext well, but not always. (Still at least 10x better than chatgpt 2 years ago). A cheet sheet in static tw5-com website will help it less wrong.

In the future, maybe only most complex UI feature will be delivered by plugins, like tw5-graph and WYSIWYG editor, or multi-column layouts.

Most of use case could be generated on user-side on the fly, and discard after use.
ReGenerate is just costing half minute of prompt.

[syrte]

Sounds like a good idea.

At the moment I’m using:

• https://tiddlywiki.com/alltiddlers.html
• https://tiddlywiki.com/dev/alltiddlers.html

I export these pages to PDF and feed them into Google NotebookLM, then ask the AI to generate the filters I need based on the documentation.

Having a more explicit entry point for web crawlers would definitely help. It could make general-purpose AIs like ChatGPT or Gemini understand TiddlyWiki better in the future, hopefully.

The challenge right now is that discussions and resources around TiddlyWiki are still relatively limited, so AI models don’t perform as well as they do on more widely used projects or languages. That’s why NotebookLM works better in this case—it sticks closely to the specific resources I provide.