AI vs. rw-designer

Published by Vlasta about 16 hours ago.

I've been thinking about the future of rw-designer in this age of AI, and I'd like to ask for your input before starting working on something that could do more harm than good.

Over the years, the site has grown into a mix of libraries, forums, manuals, and personal pages. That's great and I did not expect this those 20+ years ago, there was no big plan and things were added and changed more or less randomly, which means many things are a bit scattered. There's a lot of valuable knowledge and unique creative works here, but sometimes it's hard to find. The exact cursor set you are looking for is buried among the thousand others and it lacks a proper description and tags and so is almost invisible; similarly for forum threads - they don't always connect to each other as much as they could and there are some duplicates.

Recently, I've been exploring whether some carefully used AI tools could help improve things. Since I am not a big-tech with deep pockets, this would mean running a small AI on my PC once or twice a day (abusing my aging NVIDIA 3060) or making use of those free promotional tiers of AI services offered by OpenAI, Google, and others. This would definitely not be yet another useless chatbot capable of telling you to turn it off and on again, an automated slop generator, or anything that gets in your way.

So, before I go any further, I'd really like to hear what you think.

What kind of improvements am I considering?

Here are a few ideas, just to give you a sense of direction, some of these may even be unfeasible to actually do, but let's keep them here as they may inspire you:

Making forum threads easier to read

Some threads contain great answers, but you have to scroll through a lot to find them. One idea is to automatically generate a short summary at the top or in a tooltip:

• what the problem was
• what the solution is
• key steps

Think of it like a TL;DR that helps you do decide whether this thread is interesting for you or not.

Helping you find existing answers faster

When writing a new forum post, the site could suggest similar discussions and briefly show what the answers were.
Not to stop you from posting, but maybe save you time if the solution already exists. This may be too much as is requires instant AI processing instead of the more manageable "nightly run".

Better tagging and organization

Content (mainly libraries, possibly forum threads) could get tags fixes. Many sets have none or too many tags and sometimes irrelevant tags. Correctly assigned tags can vital for populating the "See also" section properly.

The system could over time get through old sets with no tags or too many tags and try to fix it. This may be pretty hard as it would require to use an AI that could actually recognize what is on a small 32x32 pixels image, which may very well be beyond the capabilities of current AI, especially those available to me.

So, this is a risky topic as I may spend days trying to get this to work and failing. But if it worked, it could potentially also generate new tag candidates that would be useful to add. And they will be based on already existing content. In an ideal case, this could catch trends and if a particular theme of sets (based on a new game or another cultural phenomenon) starts appearing, a new tag will eventually be auto-created and assigned to the sets.

Small quality-of-life improvements in tools and uploads

For example:

• pointing out potential issues when uploading a cursor
• highlighting things that might not work well in practice

Currently you get bullet points highlighting some common problems (not enough cursors for full set, missing descriptions, ...) But these are either auto-assigned or I have to assign them manually and I tend to be lazy. The AI could possibly add more nuance here - it may be able to catch some common errors like wrong hot spots, low contrast, who knows what else... Again, this depends on how good job an AI can do recognizing what is on a small image.

Review/comment up/downvoting

What I personally want most is to be able to not to be forced to read every comment posted and decide whether it is worth keeping (that 🙂🙂🙂🙂🙂🙂🙂🙂🙂🙂🙂 comment posted for hundredth time) or not, if I stop doing this for more than a week the comment quality gets down quickly as one spammy 67 comment just inspires dozen others to follow up.

The up/downvoting was added as an attempt to address this issue, but few people are currently using it, so I still have to skim over all the comments.

An AI could look at the comments and give them +1/-1 if they are helpful/inappropriate. This has a tiny effect, but it could serve as an indicator of which comments are appreciated and may encourage/discourage people in following the + marked comments and not the minuses.

TOTM management

I am currently picking the Theme of the Month ...uhm... themes. Sometimes, it is hard to pick a theme and do it in time (like this month, apparently). It may be possible that AI could do an OK job here, it could look at recently added sets and if multiple authors do sets that have something in common, pick that theme - it would do a contest that people are already interested in; it could possibly read a dedicated forum thread, where people would suggest themes and take them into consideration.

Why I'm asking you

This site has always worked because of its users. So it makes sense to ask before changing how things work, especially when this is controversial topic and I mostly see the web from an admin perspective and not a regular art creator or art consumer (those, who download icons/cursor) perspective.

I'd really like to know:

• Which parts of the site feel the most outdated or frustrating?
• Where would even a dumb AI actually help you?
• Where should things should stay exactly as they are?
• Do any of the ideas above sound useful or completely wrong?
• What other ideas do you have? Preferably ones that a dumb AI can accomplish, do mention even those ambitious ones, perhaps if a few years even those could be realistic to do.

Even a short comment like "please don't touch X" is helpful.