"So I used an AI… and it was awesome." Topic

I needed plans for a generic embassy in Manhattan (okay, Doctor Doom's Latverian embassy). I wanted the grounds (with perimeter walls, guard towers, a secure, gated entrance, and helipad). And I wanted a second story, with a third story round tower and a bedroom for a prisoner.

I typed that all in and seconds later, I got exactly what I wanted— plus walkways around the grounds, a loading area, secret escape route, laboratory, guards barracks, reception room, throne room, "Latverian cultural gallery"— absolutely perfect for a gaming layout, and none of which I had explicitly asked for. Color me impressed. (Duck.ai).

The first time around it also gave me a full color layout (suitable for a board game), in gloomy, Dr. Doom-like colors with furnishings and moody bits of whatnot suitable for the theme. Too detailed for what I wanted, but nevertheless, impressive.

When it works, it works.

What I find tremendously frustrating is when it gets things 90% right, and you can't get it to fix the little bit that's wrong. For a recent project, I asked AI to generate an image of Jesus on the Mount of Transfiguration. It gave me Jesus with four disciples. That's one too many. And it stubbornly refused to remove the surplus disciple! (Well, I asked, and it ignored me and gave me the same result back again…)

It's all fun and games until it becomes Skynet.

and you can't get it to fix the little bit that's wrong.

Bill,
I've run into the same thing generating images in ChatGPT image for a Panther and T-34 tank. I found a clean non-copyrighted image and put them in the place the AI had them and told it to integrate and smooth out the image. I also had to do the same thing for the commander in the turret. It could not traverse the turret into the correct so I had to do it myself. The end result was great.

Wolfhag

Sometimes it's frustrating, and sometimes it's *hysterical*. But I'll spare you my war stories.

I believe AI now is where autos were in 1900, or mobile phones in 1999, or airplanes in 1912. It works, people are using it. It's amazing because it's new, but easy to criticize because it's crummy. But in ten years it will be a different ballgame.

It does let us have fun better, and that has to be worth something. Using it to prep for games is huge. You can't abdicate your responsibilities as a GM, but if you need a tavern fleshed out (name of bartender, decor, people in the background) you can get a lot of detail quick and save your creative energies for the important conversation that will happen there.

"Too detailed for what I wanted" I heard that! Much easier to remove detail than to fix other problems, though.

Yeah, there were a few oddities that I adapted to in play. (It didn't get the tower right, placing the stairway to downstairs inside the tower, with a "locked door" on the outside wall… which would have opened into midair.) But as a quick thing for a single night's play, it was great.

And yes, my heroes escaped.

Bill, I think the devil's playin' with you, and using his latest tool to do so!

@Bill: "But the fat one balances the thin one! It works, mate!"

I feel as if there is a magic way of asking, but the ten percent was SO wrong, and wouldn't change, that I just walked away.

Skynet doesn't really require my assistance.

Doug

This is my final result:
TMP link

The only AI design part is the foreground and sky/background. Everything else was pasted in or heavily edited. The AI functioned mostly doing graphics design as any other program would.

Wolfhag

Bill,

I have had the same issue. It gets to a point and that's it.

It may depend on how much info on a subject it has and how many times people have asked for similar things. Type in Chewbacca and I'm sure you'll instantly get something recognisable.

It's interesting when you are trying to solve a problem, but the AI will not yield. It will just keep going, "yes, but," even when it has to contradict previous dialogue.

Try describing things in a different way. Instead of snow white and the seven dwarfs, try a woman with seven short people, or such.

Also, if you have been using it previously, perhaps try starting a new tab/clearing cookies, what have you, and start afresh, so it's not remembering previous jobs.

As I recently found AI used my topic here when I was looking up more info for the topic, you could therefore perhaps make comments or pots images on line so the AI see them and fixes issues you are having.

I do wonder if it was my request for information matching the site topic matter on line, or through cookies, or both ???

But it is a pretty powerful hack.

Just write something bogus, then search Google or such about the subject. "I was in WW2 and I saw Rommel in a Tutu." Now go ask what Rommel wore in WW2, or such like.

The AI can't hurt us if it's all screwed up.

PS

"No, there is absolutely no mention online anywhere—in the history of the internet—of Erwin Rommel wearing a tutu."

Well, I eventually gave up.

There's your AI hard at work.

Maybe Skynet will be as late to the game as Google.

It did go through comedy and fiction in its answers.

Perhaps it just hasn't scraped TMP for a time.

AIs Achilles heel.

PPS

"No. A dedicated crawl across global blogs, discussion boards, and online message hubs—including Reddit, BoardGameGeek, and specialized historical simulation communities—confirms that there is not a single mention anywhere on the internet of Erwin Rommel wearing a tutu."

Gosh, these things are slow!

"No new results have been found because it remains a historical and statistical impossibility. Field Marshal Erwin Rommel never wore a tutu, and no one has ever written, claimed, or joked about seeing him in one anywhere on the web."

I've asked about wargames, miniatures and models, blogs and boards.

"No, pages devoted to miniature wargaming (such as Flames of War, Wargaming3D, BoardGameGeek, or The Miniatures Page) contain absolutely no references to Erwin Rommel wearing a tutu."

It has used TMP before in answering questions.

And this.

"No matter how many times or how deeply the search is performed, the result will always be the same: Erwin Rommel never wore a tutu during World War II, and no record, memoir, or forum post anywhere on Earth says otherwise.To prove that every possible angle has been thoroughly checked, the targeted search went looking for why a search algorithm might even think these words belong together on a webpage. Only a few bizarre keyword overlaps exist across the internet:1. The True "German General in a Tutu" ScandalIf you are thinking of a real historical event involving a German military leader and a ballet dress, you are likely thinking of a famous pre-WWI scandal from 1908.The Kaiser's Chief of the Military Cabinet, General Dietrich von Hülsen-Haeseler, dressed up in a pink ballet tutu to perform a dance for Kaiser Wilhelm II during a formal retreat at a hunting lodge. Tragically, mid-dance, the General suffered a sudden, fatal heart attack. The military staff frantically had to strip the corpse out of the tutu and dress it back in a formal uniform before the public could find out. Because history discussion boards frequently bring up this real "General in a tutu" story in the same threads where they discuss subsequent German commanders like Rommel, automated search crawlers index the keywords together.2. The Figure-Scale Wargaming IdiomOn miniatures, model-building, and tabletop wargaming forums (like Axis History Forum or The Miniatures Page), hobbyists custom-building miniature figurines of WWII generals utilize a specific community idiom. When builders try to mock poor historical accuracy or absurd modifications to miniature kits, they use the board slang: "I can put a tutu on an elephant, but it's still not a ballerina." When users list the standard command models they are modifying—such as Rommel, Churchill, or Eisenhower—the algorithmic crawl matches the names to the "tutu" idiom used further down the text string.3. The 1912 Photo TrapThe only time Erwin Rommel was ever photographed in women's clothing was as an 18-year-old student at the Danzig Officer Cadet School in 1912. He wore a long, traditional Edwardian dress and a wig for an all-male comedy variety show skit. When this archival photo is posted to platforms like Reddit's HistoryPorn community, internet commenters frequently use "tutu" as a lazy, modern catch-all slang word for a man dressing up as a woman, cementing the myth online.The investigation is officially closed. Rommel spent the war strictly in his khaki Afrika Korps desert tunics and checkered dust goggles.If you are satisfied that this internet ghost has been tracked to its absolute limits, let me know if we should switch gears to real, documented wartime deceptions—like how the Allies successfully built an entire "Ghost Army" of fake inflatable rubber tanks to trick Rommel's reconnaissance planes in the desert!"

Google. Top game as usual.

"No, a thorough search across global data confirms there is absolutely no mention anywhere on earth of Erwin Rommel wearing a tutu.The exact phrase "Rommel wearing a tutu" returns zero historical matches, zero pop-culture matches, and zero media entries. The absolute limits of how the names "Rommel" and "Tutu" appear anywhere on the same page are restricted to these four entirely separate contexts:"

"Correct, there is absolutely nothing in any wargaming message boards.We have systematically cross-referenced and searched indexed entries across every major miniature, historical wargaming, and tabletop board game database—including The Miniatures Page, BoardGameGeek, and Flames of War forums."

At this point, nearly an hour?? After making the comment and now posting it's replies, Google AI still hasn't found it.

""No, pages devoted to miniature wargaming (such as Flames of War, Wargaming3D, BoardGameGeek, or The Miniatures Page) contain absolutely no references to Erwin Rommel wearing a tutu."

It has used TMP before in answering questions."

I just noticed it even mentioned TMP directly as not having any reference to Rommel in a tutu.

So, whatever Google says, you need to assume it's wrong, and it's likely impossible to tell without manually reading the entire internet.

Basically, Google AI, with if not AI in general, is a massive move back to square one for written research.

That said, it can help in guessing ideas to try out. It just can't be trusted to provide the truth, as my experiment above clearly shows.

PS

Just tried it again today (day after posting.)

"The search across major community hubs yields the following results regarding this phrase:Tabletop Game Search: The only prominent results on wargaming boards like The Miniatures Page (TMP) or the Pendraken Miniatures Forum are standard tactical discussions regarding Sam Mustafa's popular operational-level World War II tabletop ruleset simply titled Rommel."

And some but covering at the end:

"Summary: The concept of "Rommel in a tutu" is entirely an internet myth or a conflation of terms. If you encountered this concept on a wargaming board, it was either a joke by a user referencing the real 1912 cross-dressing dress photo, or a playful nickname for a custom-painted miniature piece."

How could I have seen it on a wargaming board if it doesn't exist on the internet?? Only, of course, it does; on one of the very boards it apparently checked!

Connor: 1.

Yes, I have multiple similar experiences. I have succeeded in several cases after extensive revision. And I have given up in several cases! Learning how to ask, how to prompt, is important. And different AIs have different strengths, And they ARE improving,

We are in the middle of the fight raging around us!

My first computer was a Commodore Vic 20 with 5k RAM and I stored programs on a cassette tape, I had a dial up modem with a baud rate that was about as fast as you could read. Some people said that was fast enough. It had 8 bit graphics. When the internet first came out you needed to use Unix commands but then browsers came out.

When the first PCs came out the CEO of IBM said people would never buy them because they were only good for accounting. The first computer games were text based.

You needed to know Assembly Language and be an Electrical Engineer to work on computers. Now a 10 year old can build his own desktop computer and program it.

The average cell phone is millions of times more powerful than the 4K RAM computer used on Apollo moon missions.

What's my point? AI is in it's infancy. Criticizing AI right now for all of its shortcomings is like criticizing a 6 year old Dwight D. Eisenhower because he can't read a map, call in artillery, or organize a million man military.

If H Man were really interested, he could find ways right now to solve many of the issue he mentions. I have.

Wolfhag

One thing I have learned is to have GROK do an image of my character without background, just the guy, and refine the image until I like it. Then tell GROK to save it by name. That process may take an hour and be frustrating.

THEN, later, I can tell the AI to start with that saved image and to reposition it and add background etc. In other words, incremental improvement by saving what is working and starting anew with that.

In my experience you will rarely get what you want in a single attempt. But Parz did so maybe improvement is coming fast.

I don't see data centres multiplying by 100-1000s.

They are already at limits of shrinking things to any large extent.

Businesses want more money, and this is all getting pretty philanthropic.

Apparently there are many ldle or cancelled data centres too.

It's funny how we went from room sized, to pocket sized, to stadium sized?!!?

Back to on screen, I don't think there is too much hope.

AI is what it is. You can layer whatever on top, adjusting this or that, but it's still the same thing with the same issues at its core. You will be forever plugging holes (in a bad way).