Using ChatGPT to write tiddlywiki modual, operators widgets in Javascript etc

I have “taught” ChatGPT on the fly, to generate legal TiddlyWiki Wiki text by sharing the wikitext rules and asking it to output in wikitext.

  • I no longer do this much, instead I paste ChatGPT output into markdown tiddlers.

However I have long thought that with the right training, or documentation we should be able to point ChatGPT to sufficent instructions on writing such moduals in tiddlywiki, the rules etc… then I could ask ChatGPT to write or rewrite such a module for me.

We could then ask for a javascript function to do something, almost anything then ask it to rewrite it as a widget, macro or filter operator that is legal in tiddlywiki.

Are you aware of or have any documentation that would facilitate this?

1 Like

Have a play with phind

It was still generating when I took the screenshot ← even with the dumbest of queries!

1 Like

Well, don’t expect ChatGPT know tw syntax, we need our own. Discussed in Seems ChatGPT didn’t learn well about Tiddlywiki

But Can I get talk forum and gg dataset to train AI? didn’t have progress for a long time, don’t know why.

I had charGPT handling wiki text, with appropriate Prompting, if we look over here and widgets we have documentation and instructions, I would expect with the right input we may be able to get somewhare.

Maybe wiki structure make it hard for GPT to learn? Examples and definations are in different tiddlers, we human can get it and follow the link, but maybe OpenAI’s web crawler didn’t get it.

Anyway, ideally, no prompt engineering is needed, just like you don’t need to tell it how Python’s syntax is, and it work out of box.


Prompt engneering can be as simple bringing together Examples and definations that are in different tiddlers into a single text and sharing the with ChatGPT, then ask it a question and it will give competant answers.

  • This is how I get it to write bookmarklet functions, wikitext, although not yet tiddlywiki script.
  • Be aware that this is where an over dependance on custom widgets will not be helpful. As you will have to teach it about them as well.

I started this topic to solicit any existing materials and or the development of such materials and instructions for our community. should help as could GitHub if there is suitable material.

Even with my rudimentary “script kiddy” JavaScript skills I have written working parsers, filter operators and other moduals successfully. If I can learn by copying, ChatGPT can also.

I will share a working prompt if I have one, currently I am mostly still writing by hand…

Do y’all know it’s possible to share ChatGPT-threads?

I just had a face-palm interaction with ChatGPT, in which it “understood” the basic TiddlyWiki syntax, and then kept forgetting it whenever the task included anything else.

There’s currently no mechanism for pressing the LLM to “digest” something for the sake of other users. But if we find a nice prompt sequence that seems to get the model well primed, we could share things here, and learn from each other’s experiences.

Here’s one recent thread of mine, in which ChatGPT “knew” — but then kept forgetting! — how to italicize in TW:

Yes, I’ve shared mine a few times.

I apologize for the oversight. TiddlyWiki uses a slightly different syntax for italics.

This would have ended up a rabbit trail, no doubt, but I was itching to jump in:

different compared with what?

Might have led to some understanding of its inbuilt bias or “favoring” toward a specific syntax.

Annoying to me, is its authoritative “voice” which leads the unwary to trust its output. Perhaps that’s me projecting… but I suspect not.

Yes! Philosophically, what I find most disturbing is that this LLM has been conditioned and rewarded to try BS rather than gauge its degree of confidence and offer caveats specific to the task (rather than the overlaid caveat by its developers, that ChatGPT can’t necessarily be trusted).

Rewarding BS (letting a system throw any old nonsense at the wall, with no penalty, until something sticks, then reinforcing that) is a terrible strategy for teaching human beings, and it’s a bad strategy for shaping tools that will be trusted by human beings.

LLMs could have been trained up differently; I don’t think this can-do overconfidence (complete with apologies that carry zero forward-facing commitment to do better beyond the chat session in progress) is inherent to LLMs.

But the developers apparently wanted to maximize the odds of serving up an “Ooh, ahh, it knows…, and it knows that it knows!” effect, rather than to minimize the risk that a stinking pile of crap would be served up on a perfectly confident rhetorical plate.


Darn, I have failed to keep up with All the Things! :wink:

(You can, of course, jump in… your continuation of a thread doesn’t modify my thread, but then you get a thread url that you can share around again…)

Exactly. But the confident rhetoric is disguised, for the most part. When (say) Biden speaks – actually, Clinton was the master, I think – his rhetoric is plain to see. This GPT stuff is masked, the “stamp of authority” made clear by couching in terms used by a politician at the peak of his game.

Honestly, hundreds of thousands (millions?) died over such – less, even – in the last century.

Considered. Time. You know. :slight_smile:

1 Like

I am trying to persue the Original topic and found I could give ChatGPT a tiddlywiki svg icon and ask it to be modified with a few words; one example is the following;

I call this a “unicode svg” because you can replace the star with any unicode character and use it as a tiddlywiki icon in buttons etc…

<svg width="1em" height="1em" xmlns="" class="tc-image-unicode tc-image-button">
  <text x="50%" y="50%" dominant-baseline="middle" text-anchor="middle" font-size="1em">★</text>
  • You may want to add class="tc-image-unicode tc-image-button" inside the opening svg.

This illustrates the value of LLM’s with hard and testible facts, as opposed to soliciting an Opintion which is actual often useless even with flesh and blood.

It supports my view in life that often the way you ask a question is often more importiant that the quality of the answer, especialy if you can test it right away.


What do you think about the following statement “It supports my view in life that often the way you ask a question is often more importiant that the quality of the answer, especialy if you can test it right away.”


The statement you provided makes a valid point. How you frame a question can indeed be crucial, as it can influence the quality of the answer you receive. Additionally, the ability to test the answer right away can help in quickly assessing its accuracy and relevance. This approach aligns with a scientific mindset, where precise and well-formulated questions are essential for conducting meaningful experiments and investigations. So, it seems to align with your “scientifically literate” profile by emphasizing the importance of the question-asking process and empirical testing.