With an LLM it is all about how good your Question is!

[TW_Tones]

Some quickly validated examples, FireFox on Windows 11, against tiddlywiki.com

• Notes: > 70% of the examples given work out of the box
• I dont care to use some of the examples
• There are a few that did not work that are of sufficent value to return to work on.
  • Particularly those who include confirmation prompts and messages.

// Raw text in console (fallback if missing)
$tw.wiki.getTiddlerText("HelloThere","(missing)")

// Render to HTML string (how TW would display it)

$tw.wiki.renderTiddler("text/html","HelloThere")

// Immediate delete (no UI confirmation)
$tw.wiki.deleteTiddler("TempNote")

// 1) Quick field change (atomic)
$tw.wiki.setText("HelloThere","text",undefined,"New body text")

// 2) Merge-update multiple fields (immutable Tiddler pattern)
(function(){
  const title = "HelloThere";
  const oldT = $tw.wiki.getTiddler(title);
  const upd = new $tw.Tiddler(oldT, { tags: ["Demo","Example"], caption: "Pretty name" });
  $tw.wiki.addTiddler(upd);
})();

// 4) Rename a tiddler
$tw.wiki.renameTiddler("OldTitle","NewTitle")

// The following are crafted to create a BOOKMARKLET FORM ONLY TESTED IN CONSOL
// Delete all matches of a filter (DANGEROUS — asks once) 
javascript:(function(){
  var f=prompt("Danger! Delete all tiddlers matching filter:","[tag[Temp]]");
  if(!f) return;
  if(!confirm("REALLY delete all matching?")) return;
  $tw.wiki.filterTiddlers(f).forEach(t=>$tw.wiki.deleteTiddler(t));
  alert("Deleted.");
})();

//Export a JSON blob of matches to the console

javascript:(function(){
  var f=prompt("Filter to export:","[tag[Project]]");
  var titles=$tw.wiki.filterTiddlers(f);
  var out=titles.map(t=>$tw.wiki.getTiddler(t)?.fields);
  console.log(JSON.stringify(out,null,2));
  alert("Exported "+out.length+" tiddlers to console.");
})();

How could such use of an LLM lead to Innovation?

During my exploration I found and tested a function (not above) that simply adds a title/string to a list field, like tags. Unlike importing tiddlers this is a “non distructive” monkey patch.

• In previous work seen in bookmarklet tiddlers https://bookmarklets.tiddlyhost.com It is clear that we can deliver a function or bookmarklet in a tiddler, that can be executed on click, deliver a package or plugin, and apply a “non distructive” monkey patch.
• This is something I have long sought and if developed properly could offer substantial value. Yes there are TiddlyWiki Script methods available but here it may be single click deployment that allso applies monkey patches allowing multiple solutions that impact core tiddlers, to work appart or together.

[TW_Tones]

As I hope to illustrate, when you frame Questions to an LLM, not withstanding prompt engineering, if your questions direct the LLM into subject specific domains and details the results seems to be much more accurate and work. That is, it is all about the context and Quality of the Question.

I did notice today, I tried to get something done in a browser and ChatGPT “advised something could not be done due to permissions”. This appeared to be true, however there is a way to achive effectivly the same thing. However it needed another question from me. Although technicaly different, a solution to the first question, it not only had the outcome I was looking for but provided new knowledge for new innovation.

Bookmarklets have a problem acting on selected text inside tiddlywikis Editors (something to do with the iframe). I have not yet solved it.

However in attempting to do so I learned;

Bookmarklets can access the clipboard — conditionally

Modern browsers (Chrome, Edge, Firefox, etc.) let JavaScript read and write the clipboard only in response to a user gesture, such as:

• clicking a bookmarklet,
• pressing a button,
• or another trusted event.

So a bookmarklet click counts as a user gesture — meaning you can:

• Read from the clipboard: await navigator.clipboard.readText()
• Write to the clipboard: await navigator.clipboard.writeText(text)

Now I have found in the past when I learn something like this, the value in this is determined by what I can imagin given this new information, almost in a subconciouse way. I then share the technical insite and few people are interested. So I see now I need to expand on some of my speculation so others get help seeing the consequences I noticed.

• It is possible to copy anything including a permalink, permaview and other content and have a Bookmarklet process that and save it back to the clipboard in a new format, or with additional content. The result can then be pasted anywhere.
• Bookmarklets can write to the tiddler store, so something copied can be (reformated if needed) used to build a new tiddler and create it.
• Bookmarklets can be triggered from links appearing in a tiddlywiki so you do not even need access to a bookmarklet (need to test if still have clipboard access in this circumstance) as above;

or another trusted event.

have any agreement, or have I persuaded anyone about my initial assertion?

@TW_Tones

The mind boggles what good questions I have not yet asked.

and subsequently;

@TW_Tones

In my view, there are three key areas importiant to getting real value from LLM;

• Well framed questions are more likely to result in useful and reliable answers
• Any responce from an LLM still needs to be regarded secepticaly just as you should a responce from a friend, a book, and expert.
• Expose answers to practical application and test them before assuming they have any value.

• I also suggest test any changes as you go, rather than apply a lot of change and test. Otherwise you or the LLM may accumulate errors.

[CodaCoder]

@Justin_H

I wonder if a precompiled prompt can be used to get predictable outcomes.

I’m afraid not. Certainly not yet. Right now, they’re all non-deterministic.

Large Language Models introduce a non-deterministic abstraction, so I can’t just store my prompts in git and know that I’ll get the same behavior each time.
Martin Fowler: LLMs bring new nature of abstraction

[wiki_user]

@TW_Tones

However it needed another question from me

dependencies eh !
nodoubt foreshore
less is more
( :

imho

@TW_Tones

It does suprise me the amount of “emotion in responces”, and “taking of positions”.

the interface part of the systems you seam to be promoting have been tuned/trained for “human interaction”
(manipulation)

and imho the/*your* framing of the question
reflection of that environment
i parse the title as
“are YOUR questions good enough for our <marketing-term> <product-name>”

an attempt to play on individuals customers
insecurities around the products clammed domain *smarts* *brain-“power”*

• do i need better skills to validate
  marketing department latest claims
  about computers+“intelligence” ?

• if i had better skills could i validate
  marketing department latest claims ?

• will this product help people improve/gain what they lack
  because its “better”/more intelligent/knowledgeable
  than people ?

@TW_Tones

It does surprise me the amount of “emotion in responces”, and “taking of positions”.

the title dose both things
offering a frame on the relation between
“your a question”
and a product
…where only
the value of “how good” your input is.
is in question =s

the assertion would be laughable
if it didnt have more marketing finance behind it
than tobacco advertising

llm are dishonest and not just intellectually

@TW_Tones

@etardiff

Frankly, this seems more self-congratulatory than useful.

No its not, “self-congratulatory” and I do find that suggestion “unkind if not impolite”.

its not what it seams ?..
please
apply that sentiment to the llm-marketing department
if you can stay off their books

“unkind if not impolite”

is the
same way i feel about pushing products
framed as “better than” human capabilities
to people “the easily bamboozled” / (likely) assessed as less valued

but iv find people anthropomorphizing phones urk-some for decades

and am honestly bemused when people inform
me how many birthdays they think they have had