All AI responses regarding TiddlyWiki knowledge are incorrect.
Therefore, I recommend this AI. Its most notable feature is that it does not hallucinate. However, it requires about half an hour to repeatedly gather information and verify it. Use this AI when you truly cannot find an answer.
Thanks tomzheng… its a good recommendation… I tried it out to see if Tiddlywiki has a Cooking Recipe plugin that supports the standard data format (json-ld ie Schema.org Recipe)
It was a surprise to watch it do multiple searches and report these to the UI for an extended time - some of the links seemed not relevant … But at the end it did come up with what looks like a good answer ( there is a plugin for recipes - just not with the schema.org data format ) … and it gave its references etc … so really cool.
Given how specialised the Tiddlywiki knowledge domain is - is there any thought among the community about how to develop an AI platform that has the RAG capability for searching through our knowledge articles ?
I asked this thing a question about geopolitics (I know, I know, that was below the belt) and after reading the answer, I don’t want to talk to it about TiddlyWiki for a while 
Thanks for sharing @tomzheng it is an interesting approach, and an option to consider
Such broad generalisations are always incorrect, the are plenty of good responses from LLM’s that create usable code (not with tiddlywiki script, but Javascript) but as has being discussed earlier at length here in talk.tiddlywiki, as I pointed out here With an LLM it is all about how good your Question is! and elsewhere.
The more directed and informed your questions, and the information given or suggested to an LLM the better the answers will be. What I suggest is using LLM’s (A.I. Does not exist yet) in an area of your own expertise to study and test how to use it, before applying it to a subject area you may be unfamiliar with, so you can become aware of its limitations. Before being fooled by its false certainty.
Numerous repeated check points, where you test and validate the output of an LLM in the real world is also another way to get real value in the long run.
Sadly critical thinking, scepticism and evidence is lacking in a lot of LLM’s users, and results in the sharing of “AI Slop” these lessons are poorly lacking in many users of LLM’s and could be the downfall of civilisation. But this was the case even before LLM’s, its just easier than ever to be a “weak thinker”.
I was only just listening to a “Space news podcast”, I heard one of their items about a company launching to provide hotel accommodation on the moon, when I applied a sceptical ear to the claims, I realised it was a speculative idea by a 20 something, thinking his dreams can manifest his desire to be an Astronaut (he possibly read “The Secret”) and fleece millions out of investors, intentionally or due to naivety.
Stay away from controversy
If you ask what TiddlyWiki is, AI can answer that just fine. But if you ask about specific technical details, AI struggles to get it right. If you know which AI has a high probability of answering correctly, please let me know. TiddlyWiki is a niche field with complex technology, and AI capabilities are quite limited.
ChatGPT is fine for me as long as I ask the right question, for example ask it to find all the JavaScript in TiddlyWiki.com and extract functions you can use in Bookmarklets and the console.
Or perhaps past in the code for a filter operator such as $:/core/modules/filters/compare.js then ask it to produce another with different functionality.
I’ve had relatively good results using Google’s NotebookLM (https://notebooklm.google/). You can add tiddlywiki.com site as a source, and then ask questions from there. It tries to pull answers from the source, not all over the internet. For my basic needs, it has proved reliable. I’m not sure how it would do for more complicated things things though.
I got home and tested ChatGPT.
My request was Create a fully functional layout in TiddlyWiki.
ChatGPT’s first response suggested using themes, which was completely off-track. After I refreshed the answer, ChatGPT mentioned $:/tags/Layout, but everything else was wrong. Miromind.ai got it mostly right on the first try.
NotebookLM is a decent AI.
I’m not sure if I can access this website for testing. Google AI isn’t available worldwide.
Other methods include deepwiki.
My recommendation is to use Morimind only after other methods have proven ineffective.
Thats great that;
But I do primarily use ChatGPT and have a subscription. I have being doing a lot with tiddlywiki and it is now “Aware” of tiddlywiki. I have not made the suggested layout but it seems to know what its doing.
I found that gemini pro is mostly up to date with tiddlywiki. Sometimes, I need to copy/paste the doc to improve the quality of the answers but I get pretty good results most of the time