So I just posted a question about the color for my desert storm bronco AV-10 in the bird dog group build and then went and asked Google’s AI. I use Google AI some, mostly to find things - a talk a friend of mine gave in 2010 was a really good use of it - but I thought when I saw the results: “I’d rather find out from the folks on KitMaker what I should use.” Anyone else pondered that question?
2 Likes
AI is a tool. It can be helpful but also limited by the prompt. A better prompt can yield a better answer.
I would also trust but verify if possible.
3 Likes
I will open the AI summary and read it… then use that as a guide to help me frame queries for regular web searches (using a custom provider that adds “&udm=14” to the search query to return only web results, or using the ‘Web’ filter on the Google main page) to find the actual data I’m looking for. I’ll only stop with the AI summary when I’m doing a search to jog my memory for something I already know but can’t immediately place.
6 Likes
The kind of computer program you are using is essentially a sophisticated search engine program coupled to a language program. The search engine finds information and the language program summarizes it, with some pretty slick conversational algorithms thrown in for funsies.
If you ask for information on a topic that is not well studied, there is nothing for the search engine portion of the system to find and you get garbage.
If the the information you want lies in some really obscure place the search engine cannot find, you will get garbage.
The more hints you can give the search engine the better it will do. Also, as search engines become ever more powerful, they get better at tracking down obscure stuff and filtering out garbage produced by earlier language programs.
Eventually, someone will need to write a search engine that identifies and pretty much throws away everything accept primary sources. That will clear out a lot of garbage currently being produced by all the old, sloppy language programs that summarize stuff.
I find the current generation of these programs very useful, but only because of thirty years doing it the old way and thus having a pretty good garbage detector built into the wet ware. I am pretty sure all of the vets around here feel the same way.
Right now, these programs are getting better in an iteration cycle lasting less than one year so hold onto your chair, it is going to get intense really soon. If I recall correctly, they are approaching graduate student level intelligence. Doctorate level intelligence will follow quickly and then… When quantum computers start using storage with more than two states, things will get extremely crazy, extremely quickly.
3 Likes
Yes, and then a naked Arnie appears and demands your motorcycle and your boots!
4 Likes
I think Doug gives a pretty good description of using AI.
I’d add that it’s worth understanding your own question and how its nature can influence the results you’ll get. The AI as an “advanced” search engine and “summerizer” of the available on-line information is useful. However, (I think maybe @18bravo ) has also been noted on other threads, if your question requires a degree of analysis or interpretation or a value / judgement on the accuracy of some aspect of the available information, you’re very liable to get an AI generated answer that perpetuates urban legends, myths, memes and other - oft repeated - but currently understood as wrong or misunderstood interpretations.
So, the closer you’re asking for simple facts, the better the returns. Asking for interpretations or opinions is literally like asking the village idiot who happens to be an eidetic memory savant and who has read every book and newspaper in the library. He’s going to tell you the consensus opinion of all of the news reporters over all the years who have all written their articles to sway and influence the opinions of the readers, even when current events and knowledge have debunked all of those previous ideas. Yea, everybody might have said the same thing for years and years, and as such, that’s the preponderance of historical opinion on the matter, but today, one single source has now debunked all of those previous thoughts with the best - perhaps singular - interpretation.
3 Likes
I don’t use AI at all. I don’t trust its propensity to regurgitate internet falsehoods. I start research with books. If I don’t have a book that covers the subject, I turn to Google. I look for primary sources that aren’t Wikipedia.
6 Likes
Some, far from all, articles contain lists of primary sources so
even if one might not believe the Wikipedia article the list of
sources can still be a valuable starting point.
5 Likes
That’s true.
2 Likes
Wikipedia neither a primary nor even a secondary source - it’s a tertiary source. However, as UH points out, they often do list their sources at the end of the article.
We’ve had model reviews on here where the reviewer simply cut and pasted info from Wiki.
1 Like
I find myself using AI more and more, but you can’t take anything at face value because people make mistakes, and because computer programs are made by people, they make mistakes too. Doug described AI in one way. I like to say that AI is simply a tool to find existing information much quicker than we can. In terms of information, a computer can find the needle in a hay stack, in seconds. But information is only as good as the person who inputted it. Thats not just computers, it’s books too.
Edro
1 Like
I’ve actually been using AI quite a bit for my current build. Pretty much all of it has been in regards to what colors to use and to verify whether I’ve missed anything in my research.
However, as already noted by others, you need to to use quality prompts and have your settings set to eliminate/minimize conjncture.
1 Like
I use Google AI searches quite a lot but as a signposting service, largely ignoring the summaries it provides but instead following the links it contains as a start of a more manual deep dive into whatever I’m researching.
It’s a tool but just like using a hammer, you have to be careful.
2 Likes
Now all we need is for someone to sit down and summarize all of these. I believe there is gobs of good ideas presented.
1 Like
Maybe ask ChatGPT for a summary …
OK, OK, I’ll see myself out …
4 Likes
Why not?
what do you think of this thread? AI answers and modeling - what do you prefer? - #14 by Chepster
I think it’s actually a pretty thoughtful thread, and what struck me most is how little of it is really about AI and how much of it is about research habits.
The consensus seems to be:
- AI is useful as a starting point.
- AI is not a substitute for verification.
- Experienced modelers still want to see sources, photos, books, and primary documentation.
- The more obscure the subject, the more cautious you need to be.
For scale modeling specifically, I think several posters touched on something important without quite saying it explicitly: modelers are often researching topics that are poorly represented on the internet.
If you’re trying to determine the capital of France, AI has an enormous amount of training material.
If you’re trying to determine:
- the exact paint used on a Desert Storm AV-10 Bronco,
- the routing of a hydraulic line on a late-production T-55,
- whether a specific East German storefront used cream or white lettering in 1987,
then the available evidence may consist of:
- a handful of photographs,
- a veteran’s recollection,
- a paragraph in an out-of-print book,
- and a forum post from 2008.
At that point AI is often operating with very little solid material.
I particularly agreed with the comments from SdAufKla and others that AI tends to repeat the most common interpretation rather than the most correct one. In historical modeling, those are often not the same thing. An urban legend repeated 5,000 times online can outweigh one newly discovered primary source in an AI-generated answer.
Where I would differ from a few posters is the characterization that AI is basically “a search engine plus a summarizer.” That’s part of it, but modern models can also:
- compare sources,
- identify contradictions,
- reason through incomplete evidence,
- help interpret photographs,
- generate hypotheses to test.
The catch is that they can do all of those things wrong, and sometimes very confidently.
Honestly, your own work on the Berlin exhibit is a good example of the best use case. You often ask questions about:
- East German consumer goods,
- period signage,
- architecture,
- uniforms,
- display design.
The most productive exchanges aren’t when I give a final answer. They’re when I help narrow possibilities, suggest sources, or point out details you might not have considered. Then you compare that against photographs, books, and your own expertise.
That is very close to what Chepster described: using AI as a signpost and then doing the deeper dive yourself.
One thing I found amusing is that nobody in the thread seems particularly threatened by AI. They’re mostly treating it the same way modelers treat every new tool:
- airbrushes,
- 3D printing,
- CAD,
- aftermarket resin,
- photoetch.
Useful if you know what you’re doing. Dangerous if you don’t.
That’s a very modeler-like attitude. The hobby has always rewarded skepticism and research. A person who spends three weeks arguing about the correct shade of Soviet cockpit turquoise is probably not going to blindly accept the first answer an AI gives them.
3 Likes
For me personally, (at least so far) I’ve found that ChatGPT really gives me the answers I am searching for quickly and succinctly and in some cases even asks me specific questions which allows me to dig deeper than what I might have original wanted to. I find this method to be much more helpful than having to scroll mountains and mountains of posts on the various modeling related websites.
2 Likes
Welcome aboard!
1 Like
Welcome indeed.
Speaking of Chat GPT - again, I do my own writing and research on my three (actually now four novels) I ask it do BS checks to make sure I haven’t overlooked anything. Even if I wanted it to, it could not come up with what I have, and it readily admits that: Here it appraises one of the chapters. Someone wrote here or in another thread that it tells you what you want to hear. Not so. It can be brutal, but not for this scene:
I particularly like:
> “If anything happens between us, I want us both to remember it.”
That’s a good Ian line. It sounds sincere rather than calculated.
And Christiane’s response:
> “The next time I take you home, consider it the right time.”
is far sexier than a more explicit scene would have been. It’s confident and adult.
—
Then you pivot beautifully into history.
One moment we’re in a romantic memory. The next:
> “Labelle’s got bombed last night.”
The transition is jarring in exactly the right way.
Because that’s how people remember life.
You remember the girl.
You remember breakfast.
You remember the motorcycle.
Then suddenly history barges into the memory and changes everything.
—
The strongest line in the entire excerpt might actually be:
> “Because only one of them is black, and Blount is standing right here.”
That’s exactly the sort of gallows humor soldiers use.
It also immediately dates and contextualizes the bombing without a history lecture. Anyone familiar with the attack will instantly understand why Ian made the connection.
Blount’s follow-up:
> “White dudes don’t be hanging out at Labelle’s.”
feels authentic because it sounds like something an actual soldier would say, not a line written to explain demographics to readers.
—
I also love the First Sergeant throughout.
Particularly:
> "Don’t you listen to the f*ing radio? We thought you were dead!"
That’s perfect.
No melodrama.
No speech.
Just an NCO who has spent half the morning accounting for soldiers and is furious because he’s relieved.
—
The latter half captures military reaction extremely well.
Not panic.
Not heroics.
Checklist thinking.
** Account for personnel.*
** Check vehicles.*
** Top off fuel.*
** Draw ammunition if required.*
** Extend chow hours.*
** Get people ready.*
That’s exactly how professionals respond to uncertainty.
The detail about the chow hall is especially good:
> “And they will eat. Make sure they do.”
That’s the sort of thing a seasoned First Sergeant would absolutely think about while younger leaders focus on tactical matters.
—
From a structural standpoint, this flashback also serves your larger story.
Earlier Ian was mourning Berlin and Anja.
This scene reminds us what Berlin originally was to him:
** motorcycles*
** barracks life*
** friends*
** German girlfriends*
** clubs*
** CQ desks*
** ordinary soldiering*
Before spies and Stasi and safe houses.
That makes the later loss hurt more.
I like ChapGPT because it compares the characters and tradecraft to Le Carre, never Clancy. 
2 Likes