How would a hypothetical cross-wicking bot work, specifically?

So— there's been mention of a cross-wicking bot across several different threads— im combining that into one place so it stops derailing. Question is how, specifically, would something like this work? Is it even actually a reasonable suggestion?

Some discussion starts here, and I'll summarize what's been suggested so far

---

Basic setup, suggested by ~Amonimus:

• For a select page, locate the example list. Some pages may have small lists as part of the description (installments of a franchise, performers in a band), or multiple example lists split into folders, so it would need to figure.
• Get the list of examples as an array and store in memory. This would be every line that starts with "* " or "# ". The "** " and "## " would be a part of the previous example.
• Get the main wick link, normally it's the first one. Use it as a key.
• Open the page's Related tab.
• For each key in the example list, check if there's a corresponding page in the Related list.
• Open each page that does not link back to the current page.
• Get a list of examples there the same way.
• Sort there example list there alphabetically if it's not already. If there are folders, they should be sorted independently.
• Locate where to put the held example. On a trope page it needs to find/make the folder based on work's namespace according to Media Categories FAQ. Though we can't really distinguish Western Animation series and movies without context. On a work page it's normally just one big list, but sometimes it may not have a list and have like Tropes A To M and Tropes N To Z subpages instead.
• Paste, replace the key wick with the wikiword we came from, save. If an example uses a free-form writing instead of a strict "* Trope/Work: Example.", this would however result in a grammatically incorrect sentence.

In conclusion, for every example on a page, the Related tab now shows pages corresponding to each example.

---

Current issues that need solutions:

• It would need to delete from both pages if removed, since otherwise it would be adding back in examples that someone removed from one page if they still existed on the other page. Or at least not keep adding them back.
• Work pages with enough wicks get split off into subpages. This could be any combination of:
  • Characters
  • Recap
  • Tropes [letter]-[letter]
    

If a trope is about a specific character, we should prioritize the character page, if one exists. If a trope is about a specific episode, we should put it on that recap page, if one exists.

• some examples on pages are formatted by subworks, say-
  • Star Trek has many examples
    • Example from TOS
    • Example from TNG
    • Second example from TNG
      

This shouldnt go on the franchise page as a group, it needs to go on the individual pages

Edited by Tremmor19 on Aug 1st 2023 at 5:37:09 AM

Good summary! Mentioned this on the other thread, but if we still want manual review, we could always have it go one at a time where we either approve, tweak, or skip a wick.

yea, that seems best. honestly the main thing a bot could be doing is the tedious repetitive parts— taking care of the steps where you individually click on each link, open in a new tab, click edit, find the correct location, copy and paste the example, switch the trope name to the work name, etc, for each trope on the list. Its no trouble at all to manually confirm that it worked.

Edited by Tremmor19 on Aug 1st 2023 at 8:18:04 AM

Paging ~Amonimus as pings don't work in OPs

TRS Queue | Works That Require Cleanup of Complaining | Troper Wall

It is a pretty well established fact that folks tend to not review automated suggestions overmuch (automation bias) so we can't rely on people double-checking a bot's addition. We'd still get a lot of incorrect edits that way.

I admit, way back when I was considering the idea of a crosswicking bot too but the thing is, the amount of special cases and pitfalls that would need to be programmed for is large. And the amount of programming needed to handle them all too. Possibly more than the bare effort needed for 2.0, which also offers far more functionality than a crosswicking bot. So I am not certain that it's the best way to invest development resources.

"For a successful technology, reality must take precedence over public relations, for Nature cannot be fooled." - Richard Feynman

My preference would be, if we develop a crosslinking bot, develop it for the very specific cases where it'd be most useful. So less replacing the need to crosslink manually in general, and more for crosslinking a new trope or work page by going to each page and inserting each example as needed. But as suggested even that might still pose problems.

Would an AI enabled crosswicking bot be a solution? These A.I.s are pretty good at generating well-running text from inputs, if nothing else.

Optimism is a duty.

GPT is for synthesizing new text based on a question and how people normally reply to it. I see no utility of it (or any AI) in "copypaste text to another page at the right spot" taks.

TroperWall / WikiMagic Cleanup

You're thinking of a very specific implementation of GPT, but you could just as easily get it to rewrite existing text to be grammatically correct.

Optimism is a duty.

Whomever runs the AI would be responsible for ensuring it runs "correctly".

I had a dog-themed avatar before it was cool.

But who would be running the AI in the first place? Each troper who wants to crosswick something? That seems like more work to make sure things go smoothly than it is to just do crosswicking yourself.

Currently Working On: Incorruptible Pure Pureness

No, presumably the AI would be run on the TV Tropes server.

Optimism is a duty.

i had briefly made an attempt to see if i could make a wick-check bot (trying to do this first as this would be both easier and less disruptive than making a crosswick bot) (i didn't really finish it). in the version i was working off, it would just collect all of the examples from the related tab that looked correct and print them as a text file— the user would then have to add them or sort them manually. This might have some risk of people not actually reading the examples, but at least it would still require the user to be involved.

I can't really picture how gpt would make this much easier— i certainly wouldn't want a bot that is actually adding things to the wiki instead of having a human do it. Seems very risky

i think what they meant was, who would be in charge of pushing the button and confirming that the bot did it's job correctly, not where the bot would actually be hosted. (forgive me if i misunderstood)

Edited by Tremmor19 on Nov 5th 2023 at 12:10:21 PM

Yeah, I'm okay with auto-crosswicking in the way it was planned for 2.0, but with current wiki software... Such a bot would need to:

• Identify the correct page, which would include locating specific subpages (eg, media specific pages, character pages, recap pages, etc)
• Identify the correct folder or section on a page
• Sort alphabetically using our alphabetization rules
• Rewrite the example so that it's suited for the opposite page

There's just too much room for error, as AI currently still has issues understanding exact commands and prompts.

Edit: And yeah, that's what I meant. Who would be in charge of actually making sure the examples got crosswicked? The individual user, who would either be using a wiki tool or worse, would just be stuck with an auto-AI crosswick bot that'll force them to check on every occasion.

Edited by WarJay77 on Nov 5th 2023 at 12:12:46 PM

Currently Working On: Incorruptible Pure Pureness

If someone can be more specific what an AI can do aside "well AI do a lot these days", I'm interested. But for a job that requires high percision while conditions may vary, it'd be much easier to split examples with Regex and locate where a new example would fit media-wise and alphabetically.

Edited by Amonimus on Nov 5th 2023 at 8:16:11 PM

TroperWall / WikiMagic Cleanup

yes, that's the number one thing i was discoverong when i tried to figure out how to get one to function — this place is a mess. There's just not really any practical way to get 100% accuracy on locating and swapping example types, so the only solution is to have the user manually confirm each one— it just wouldn't end up saving that much time

edit: Sandbox.Wick Check Bot is what i had started with— it kinda works but crosswicking would be 3x more complicated and not save that much time anyway, imo

Edited by Tremmor19 on Nov 5th 2023 at 12:23:52 PM

A bot for checking pages could be useful - not to actually categorize each use, but just returning a list of contextualized links that can then be categorized by humans. How it could work, in my mind, is:

• Randomly select a number of pages from the Related tab, the higher of 50 or the square root of the total listed on the Related page. The user may also specify their own preferred number.
• For each page, create a bullet point starting with a link to the page in question, using "dot" notation.
• Look for a Wiki Word to the page, or a redirect thereto, in each page's source. If it appears multiple times on the page on multiple lines, each one gets its own sub-bullet point.
• For each time the Wiki Word appears:
  • If the un-potholed link is the first thing that appears in a bullet point, copy the rest of the bullet point.
  • If it's the only thing that appears in the bullet point, attach "[bare ZCE]" to the page name or sub-bullet point.
  • If it appears in a bullet point that starts with an un-potholed link to a different page, attach "under [name of page at start of bullet point]" to the page name or at the start of the sub-bullet point, and copy the sentence where the link appears.
  • If it appears anywhere else, just copy the sentence where the link appears.
  • If it's a pothole, render the text of the pothole in boldface.
  • If it's commented-out, indicate as such next to the page name or at the start of the sub-bullet point.
• Return the results in a text box rendered in wiki markup for the user to copy-and-paste, or simply paste it directly onto a wiki page.

This may or may not be as easy to code as I make it sound, and I don't know how much would already do this, but it would probably be easier to make work if it was operated on the wiki itself and built off of existing code for e.g. the related page and the wiki parser. If it would be too much of a strain on the servers it could be for mod use only.

The main benefit of this would be to automate randomization of usage checks, which people use a variety of tactics of varying degrees of effectiveness and usefulness to do as it is. The secondary benefit would be to cut down time dedicated to the busywork of finding each link and copying over the context. You'd still need to categorize each link, and that's where most of the work comes in as it stands, but that might be all you'd need to do.

Edited by MorganWick on Nov 6th 2023 at 5:00:21 AM

I think that to avoid lazy automation bias, instead of a bot cross-wicking, we should have a second editing window open up any time we edit a page, that reminds us to cross-wick and suggests where it should go. The editor would be free to change not only the entry but also where it will be cross-wicked (such as putting it on a characters page instead of the main work page) but it will be easier and quicker than currently.

Stories don't tell us monsters exist; we knew that already. They show us that monsters can be trademarked and milked for years.

The version I have should (in theory) do something similar to but it provides the list of examples in the form of a .txt file, which the use has to add manually.

It currently splits the entire page into lines based off the presence of a line break, and then takes the lines which start with * TropeName or * {{TropeName}}, or, if it fails to find any, the lines which contain TropeName (up to a character limit, so it doesn't grab the entire description or something)

Then it outputs them in this format:

• PageItCameFrom: example text blah blah blah

It doesn't have a ZCE marker, but that's a good idea

It's currently grabbing a random selection of wicks (since its a wickcheckbot), the total number is chosen by the user. Could be pretty easily changed to just grab all of them

Edited by Tremmor19 on Nov 7th 2023 at 2:37:50 PM

This would be easier if the editor had Wikimedia-style line-by-line comparison of the preview. Then it could,

1. On Article Edit page, it would have a "Also, crosswick" button
2. When pressed, it checks current changes
3. It checks if edited lines are valid examples (has indentation and at least one wick, same how YMMV and Trivia flags are determined)
4. For targeted wicks, load corresponding page's source
5. Check if those wicks are also example wicks (same as before)
6. If an existing example
   1. If finds multiple, ask which one is related.
   2. Add another preview box for that page's current and suggested edit comparison and offer to correct manually.
7. If can't find an example, find an example list to insert into
   1. If there are multiple
      1. If it's an alphabetization soft-split, work out based on the wick
      2. If it's a medium soft-split, determine from current page's namespace. If the current is a subpage (Characters/), find the main page first.
   2. If there are no example lists: See if there are corresponding hard-splits
   3. Add another preview box for that page's current and suggested edit comparison and offer to correct manually.
8. If everything fails, suggest specifying the target page and line where to copypaste

TroperWall / WikiMagic Cleanup

I don't think that would be very user friendly. You have to make it as easy as possible for the casual, non-tech-savvy user, or else they won't do it.

A solution that looks like something a programmer might use is not good enough.

That means no generating external files for review, for one. Everything should be done within the browser, preferably within the same tab.

Edited by Redmess on Nov 8th 2023 at 10:35:13 AM

Optimism is a duty.