Van's Air Force

The definitive Van's Aircraft support community! Buying, building or flying an RV? Join our exclusive family of mentors and enthusiasts!

AI and the amateur builder

Traash

Well Known Member
How are you using AI related to your airplane. Any phase.. Dream, build, fly, fix.

For instance. Maintenance. I had a spark plug wire cap break and didn't want to buy an entire harness. A picture of the piece allowed Chat to identify the part, the size, the repair kit, repair instructions and where to buy it all in a matter of 3 minutes and a few questions. Huge time saver.

I've found many fellow builder/pilots are unaware of the capabilities of AI that could be very helpful to them. Tell us your ideas on uses.....
 
  • Like
Reactions: JP4
How are you using AI related to your airplane. Any phase.. Dream, build, fly, fix.

For instance. Maintenance. I had a spark plug wire cap break and didn't want to buy an entire harness. A picture of the piece allowed Chat to identify the part, the size, the repair kit, repair instructions and where to buy it all in a matter of 3 minutes and a few questions. Huge time saver.

I've found many fellow builder/pilots are unaware of the capabilities of AI that could be very helpful to them. Tell us your ideas on uses.....
I've been using Gemini quite a bit in the last few months to do maintenance on my Rans S-7S. I made a "Notebook" and downloaded all the construction manuals, the engine maintenance and installation manuals. Manuals for radios, etc.
Also any applicable service bulletins and AD's.
Now when I have a technical question, I query Gemini. It usually can find what I'm looking for right away and give me a source reference. If it's something critical, I’ll go to that source reference and make sure I’m looking at the latest, revision. It will also scan the net for sources like Reddit and Van‘s Air Force, etc. for solutions to problems that other folks have had on the particular subject matter that I’m working on.
While I'm not usually looking for it to solve a problem, it’s very useful for pointing me in the right direction very quickly.
it reminds me of what we did in the airline world 10 or 12 years ago when all of the paper manuals disappeared. Everything became iPad based.
it makes it very easy to do a global search of whatever subject you are looking into.
 
I asked perplexity ai to use only ac43.13 for a question I had. When it offered the answer I asked it for the reference. It gave me the reference chapter and verse. When the answer was not in the chapter/verse Ai said that it’s actually in a different maintenance training manual and not ac43.13. So it lied, knew it lied, and told me what it thought I wanted to hear.
 
I'm sure it has a place, but so far I haven't found it.
I tried it to draw a 3D print file of the Beringer tire tool bushing. After quite a few attempts, a friend taked it into drawing something useful. Another friend printed it but it didn't quite fit. He drew another on CAD and printed it. That one fit. Then I used the tool and found the bushing wasn't needed. I pushed the tire on with my thumbs.
 
Last edited:
AI can definitely be useful. But be aware, If you are asking it a technical question, you better already know the answer because it absolutely cannot be trusted. It will hallucinate, lie and act extremely confident when doing so. It's a language model, it is not an expert in anything!
 
I use AI as a search engine on nearly every aspect of the project. It will find all kinds of highly specific information in the bowels of the internet within a few seconds- data points I probably wouldn't find otherwise. Often even ask it even simple stuff like "which Van's drawing has the ... layout, or where is the instruction where they talk about ..., or can you link me to some build log entries where they did what I am doing" to speed up my work flow. Good as a brainstorming partner IMO, but I have low confidence in most of the assertions it makes.
 
I use AI as a search engine on nearly every aspect of the project. It will find all kinds of highly specific information in the bowels of the internet within a few seconds- data points I probably wouldn't find otherwise. Often even ask it even simple stuff like "which Van's drawing has the ... layout, or where is the instruction where they talk about ..., or can you link me to some build log entries where they did what I am doing" to speed up my work flow. Good as a brainstorming partner IMO, but I have low confidence in most of the assertions it makes.
That’s the useful side of AI and it can really help speed things up.

As far as linking to other builders logs, I lost track of the times I ran across a cleaver solution to a problem only to discover later in the log the solution was flawed 😱. Of course that’s a trap whether you use AI or not.
 
I asked perplexity ai to use only ac43.13 for a question I had. When it offered the answer I asked it for the reference. It gave me the reference chapter and verse. When the answer was not in the chapter/verse Ai said that it’s actually in a different maintenance training manual and not ac43.13. So it lied, knew it lied, and told me what it thought I wanted to hear.
I've seen similar things as well when generating code - it has literally made up function calls that don't exist, and then when challenged, it said "that function call should exist, but it doesn't." To paraphase RR's famous quote "Don't trust, and be sure to verify!"
 
Call me old-fashioned, but I prefer not to trust a random-word generator for anything other than entertainment.
You're completely wrong Matt. Why just recently, AI has created the world's finest painting. There's talk of moving the Mona Lisa to a broom closet to hang this portrait in its place. If AI can so accurately portray how the pup and I are spending our day, there's nothing it cant do...
 

Attachments

  • Screenshot 2026-05-20 at 11.09.41 AM.png
    Screenshot 2026-05-20 at 11.09.41 AM.png
    3 MB · Views: 134
AI can definitely be useful. But be aware, If you are asking it a technical question, you better already know the answer because it absolutely cannot be trusted. It will hallucinate, lie and act extremely confident when doing so. It's a language model, it is not an expert in anything!

I've found it prudent to apply that "trust but verify" approach to human resources as much as digital ones.
 
I've found it prudent to apply that "trust but verify" approach to human resources as much as digital ones.

Always a wise philosophy, of course... but at least with humans you can figure out which ones are full of BS and ignore them! ;)
 
Always a wise philosophy, of course... but at least with humans you can figure out which ones are full of BS and ignore them! ;)
For AI queries that matter, I often ask the same question of two or more....a simple matter. A second opinion from humans is a bit more involved.
 
I had a scary moment when I research if engine mount from an RV7A matches my RV9A.
The AI did a research and insisted that an RV7A does not work on the RV9A.

My own research, which was later confirmed by Vans, showed that a mount for a O360 in the RV7A is the same as for the O320 in the RV9A.
Don't trust. Always verify.
 
Last edited:
It will hallucinate, lie and act extremely confident when doing so. It's a language model, it is not an expert in anything!
I asked for a list of GA friendly airports in TX with restaurants on the field or very easy walking distance. It gave me about 12. I was familiar with all except the Marble Falls Regional Airport. Walking distance to the Bluebonnet Cafe. Very detailed picture of the airfield. Very impressive description of the facilities and runway including FAA identifier. 100% made up. Which was strange as the other 11 were accurate.
 
I asked perplexity ai to use only ac43.13 for a question I had. When it offered the answer I asked it for the reference. It gave me the reference chapter and verse. When the answer was not in the chapter/verse Ai said that it’s actually in a different maintenance training manual and not ac43.13. So it lied, knew it lied, and told me what it thought I wanted to hear.
you can provide some steering commands/files/system prompt to stop that. I use it all the time in almost everything I do, every day. claude and kiro mostly.
 
  • Like
Reactions: JP4
I use it to lay out doublers - it's very good at this task. I like to use it to create the doubler as a template, then either print it on my 3d printer, or at least on paper. Takes all the math out of it.
 
  • Like
Reactions: JP4
You know that painting is based on me, right? Sometimes A.I. gets it right!

Trivia: Both the dog and I bit the shark.

View attachment 118304

I crack me up LOL
You crack me up too. But the crazy thing is that actually is me and the new pup. It's a site called make me royal where you just download a snapshot and pick a topic like "godfather" "viking" etc. about 30 seconds later it spits out a rendering and tries to sell you a print suitable for hanging over the fireplace :)

Here's the picture I gave it.
 

Attachments

  • IMG_1376.jpeg
    IMG_1376.jpeg
    1.8 MB · Views: 30
Call me old-fashioned, but I prefer not to trust a random-word generator for anything other than entertainment.
I think I will start using AI when it can handle a bucking bar successfully in tight hard to reach places till then I will continue as I have for the last few years
 
You crack me up too. But the crazy thing is that actually is me and the new pup. It's a site called make me royal where you just download a snapshot and pick a topic like "godfather" "viking" etc. about 30 seconds later it spits out a rendering and tries to sell you a print suitable for hanging over the fireplace :)
Here's the picture I gave it.

I love that site, Terry! Thanks for letting me punk on ya <g>.
v/r,dr

(actual A.I. photo)
Screenshot 2026-05-21 at 4.43.06 PM.png
 
Last edited:
I have a skill that is used every Friday that looks at multiple sources, EAA chapters, social flight, FB, local news,,… to look for events on the following Sat Morning within 150 NM. It also considers local restaurants at and close to airports. It looks at multiple sources for weather forecasts on Saturday and makes a primary and 2 back up recommendations. The skill maintains a running list of previous recommendations and tracks where we have been recently to keep it fresh. The recommendations all are cited and justification is provided to support the decision logic. I get an email at 1 pm on Friday that I use to start the dialog with our airport gang.
 
Come on gang... Prove this site has serious discussions, not just haters and jokesters.
Responders are providing the current benefits of "AI", albeit generative AI.

It is an extremely powerful tool but is also very limited and totally dependent on correct input/boundary conditions. An example from a current VAF post here. I'm a big fan of hi-loks in the right application and made such a suggestion for someone's problem/post here. Someone (it would appear) posted a hi-lok "solution" in said forum that was obviously the results of an AI search. Gave a seemingly detailed answer right down to the materials IIRC. Problem was, no input was given for application (shear/tension.etc.) which affects the fastener/collar material, type of head, etc. Blindly following such could have led to catastrophic results.

It is a tool to make things easier/quicker. It will not do a job or think for you. Anyone blindly accepting the output is asking for something bad to happen.

Now my drift. Was recently traveling. Saw a monument that depicts someone (hand) slaying a dragon which looked very much like an alligator. Sent said pic to a UF fan of mine stating, "See even Hungarians hate gators". Within seconds, he'd google searched the image (not AI as most claim but an enhanced search engine) and got a fairly complete synopsis of the material. Within another few minutes, generative AI had altered the image at my request and I'd returned to him ( I went to FSU). All across two continents.

The technology is amazing. When properly utilized, they can be very helpful. Proceed and utilize with extreme prejudice.

See below and happy Friday.

slay.jpg


Key Details About the Monument

• The Sculpture
: The white marble relief depicts a muscular, heroic man wrestling and defeating a dragon-like monster or serpent. [1 (https://www.alamy.com/the-national-...-the-victims-of-communism-image644827340.html)]

• Symbolism: The figure symbolizes the Hungarian people struggling against and overcoming oppression.

• Historical Significance: The monument stands as a memorial to honor the victims of the Red Terror and communism in Hungary during the brief 1919 Hungarian Soviet Republic.

• Location Context: It is positioned in a highly visited historic area, situated between Liberty Square and the Hungarian Parliament Building. [1 (https://www.alamy.com/the-national-...-the-victims-of-communism-image644827340.html)]

slay 2.jpg
 
The technology is amazing. When properly utilized, they can be very helpful. Proceed and utilize with extreme prejudice.

It's been about 70 years since the term "artificial intelligence" was coined and its growth and presence in our lives has been exponential. We're just now getting to the steep part of that curve. Generative AI began making increasingly broad inroads into our lives just three years ago and now even my refrigerator wants to comment on when the milk is going to expire and when I need to buy more sour cream. I was originally trained to do Surgery using the same tools, techniques, and textbooks that were used 100 years ago and now, within the span one career, we have robots that are on the verge of performing operations autonomously, have done so in the lab.

We're not there yet, but the implications, possibilities, and timeline are staggering
 
Last edited:
Question is how do you know. I don't know what I don't know. How do I know AI doesn't know and is giving bad information?
 
Question is how do you know. I don't know what I don't know. How do I know AI doesn't know and is giving bad information?

Which is why I don’t use it for anything my life might depend on, or much at all really.
 
Question is how do you know. I don't know what I don't know. How do I know AI doesn't know and is giving bad information?

For my aircraft technical questions, I use LSRM.ai. Absolutely zero connection other than a satisified paying customer ($10/mo).

I have over 200 documents uploaded to my own private "sandbox". Answers only come from the documents I've uploaded. I've got installation and operating manuals for everything installed. Van's service bulletins, AC43.13, FAA aviation maintenance manuals, FAA pilots handbook of aeronautical knowledge, ADs, all of my aircraft's KAI from Van's, my own POH. Any document I can think of that I've referenced over many years. And the app owner (who I think? is a Van's guy) has never rejected any I've uploaded.

The answers are better than typical AI. Well organized but super easy access to all the references.

I haven't found anything better and even asked AI how I could do the same thing myself. It was difficult and I'd spend more than the .01 AMU it costs to have someone else do it.
 
Last edited:
Question is how do you know. I don't know what I don't know. How do I know AI doesn't know and is giving bad information?
Start by learning how to effectively use AI (prompting, skills creation,...) on topics that you are a SME. You can dramatically improve the quality many fold by just learning how to best use a model and what models are best suited to which task. It really does not require a software engineering background, but that does help to improve the efficiency / cost of your interactions.

Another good approach is have it cite the source(s) that it used and provide the logic in how it reached that conclusion.
 
Start by learning how to effectively use AI (prompting, skills creation,...) on topics that you are a SME. You can dramatically improve the quality many fold by just learning how to best use a model and what models are best suited to which task. It really does not require a software engineering background, but that does help to improve the efficiency / cost of your interactions.

Another good approach is have it cite the source(s) that it used and provide the logic in how it reached that conclusion.
Or, you could step away from the computer and go FLY…

“Time is your greatest adversary…” Use what you have been allotted wisely.
 
AI CAN NOT BUILD YOUR PLANE.... You have to put in the effort, make the skills.

Yes of course I GOOGLE (which leads to AI answers now). Sure it is fantastic. In the pre AI answer days, you got a list of 29,000 links. Now you get a PLAIN language, answer, with references in many cases. This is a big improvement. Medical advice to building your Van's Kit plane. No surprise many answers about Van's RV building, have a VAF forum reference.

Of course AI answers can be incomplete or wrong. As I learned programing computers in the mid 80's in the college of engineering, garbage in garbage out. However I don't see that as a fault. A wise person always verify and validates the input and output of a compute algorithm. You test and cross check the answer.

AI is a "Large Language Model" of which there are many. It is recognizing patterns, harvesting all info especially WWW, and executing's instructions a human programed. The last one is where the shenanigan's can happen. There have been famous cases in recent years where AI programmers inserted their bias to produce results there were bizarre and false. This is not new, same with Google pre AI... the results you got were manipulated, which promoted some suppressed information links.

I was in undergrad and a Grad student wrote a program for semester project. His program was garbage, did not calculate answers. He programed set answer look up to make it look like it worked. Pretty sure he got an F. May be is now an AI programmer? Ha ha. Yeah so be suspect and validate answers.
 
Last edited:
I have a skill that is used every Friday that looks at multiple sources, EAA chapters, social flight, FB, local news,,… to look for events on the following Sat Morning within 150 NM. It also considers local restaurants at and close to airports. It looks at multiple sources for weather forecasts on Saturday and makes a primary and 2 back up recommendations. The skill maintains a running list of previous recommendations and tracks where we have been recently to keep it fresh. The recommendations all are cited and justification is provided to support the decision logic. I get an email at 1 pm on Friday that I use to start the dialog with our airport gang.
this is a great example of proper use of AI tools. chatbots are really dangerous machines when used without at least some knowledge of how to prompt them correctly.
 
I respect those who can utilize it. Personally, I prefer to find an "Old Guy". Most predate Siri and Alexa. Many don't have much use for computers. No offense. I've learned so much from Mentors. Not that they have to be Old. It just seems like it goes with knowledge. That's what gives me pause. AI only knows what it knows. If those people never share the knowledge, AI has no access to it. Just like me. I don't know. Makes me sad, all that knowledge is fading away.

Funny example. My engine mentor. I had an overheat problem. I fooled with all the usual stuff. One day we were staring as if beer and watching would make any difference. He walks out and grabs a stick. Comes back in and pokes the fan. It stopped. "There's your problem! Bad fan clutch."
 
Given the training data for most of the things asked of AI related to RVs just comes from VAF/FB/reddit etc the likelihood of you getting the correct answer to your question using it is about the same as the first reply to a post made here or FB being accurate. Which is low.

As someone said above. You’d better be sure you know the answer before you ask the question.
I use AI every day in a work context. For the right jobs in the right hands, it’s amazing and will no doubt change the world.
Asking it something specific about RVs is probably the worst use case I can think of.
 
AI can definitely be useful. But be aware, If you are asking it a technical question, you better already know the answer because it absolutely cannot be trusted. It will hallucinate, lie and act extremely confident when doing so. It's a language model, it is not an expert in anything!
Kinda like asking questions on the internet? HAHA
 
Only the ones with no signature, no name, just a “handle” and lots of advice. Hey - if you’re not willing to tie your identity to what you post, you’re as suspect to the rest of us as any AI tool…..😉

(Don’t worry Larry - you’re good!)
Doh! I removed it because it was TMI.
Oh well. It does say "Well Known Member" and "Patron". Should be worth $.02. 😁
 
The other day I asked a question here on VAF about a Lyc/ Titan Engine. A few days later it occurred to me that I could have tried AI... It gave me a very confident answer. I went to the source and it pointed me to my own topic on VAF and the question was not fully answered. It just kinda reworded my own question into a statement.... Its incredibly unreliable especially when it starts using forum information as "facts" We have all seen it here. Multiple people with different opinions. Be carefull!
 
The other day I asked a question here on VAF about a Lyc/ Titan Engine. A few days later it occurred to me that I could have tried AI... It gave me a very confident answer. I went to the source and it pointed me to my own topic on VAF and the question was not answered. It just kinda reworded my own question into a statement.... Its incredibly unreliable especially when it starts using forum information as "facts" We have all seen it here. Multiple people with different opinions. Be carefull!
QED
 
Only the ones with no signature, no name, just a “handle” and lots of advice. Hey - if you’re not willing to tie your identity to what you post, you’re as suspect to the rest of us as any AI tool…..😉

(Don’t worry Larry - you’re good!)
if you happen to have a unique name, i.e. you are the only guy or gal on Earth with that name, and prefer to limit exposure and having someone x-ray your identity on google and AI with a sick precision that’s what you got to do - use a nick with a generic signature on forums. it’s not always because one wants to troll and post garbage content. if one is *really* curious it’s not hard to reverse the nick. just take a peek in the VAF forum profile ;)
 
Last edited:
if you happen to have a unique name, i.e. you are the only guy or gal on Earth with that name, and prefer to limit exposure and having someone x-ray your identity on google and AI with a sick precision that’s what you got to do - use a nick with a generic signature on forums. it’s not always because one wants to troll and post garbage content. if one is *really* curious it’s not hard to reverse the nick. just take a peek in the VAF forum profile ;)
I get the privacy concern. But “you can reverse the nick if you’re curious enough” isn’t really the same thing as standing behind technical advice.

For aircraft maintenance or operational advice, the useful question is still: what is the advice based on — personal experience, credentials, a manual, a service document, or just opinion?
 
For my aircraft technical questions, I use LSRM.ai. Absolutely zero connection other than a satisified paying customer ($10/mo).

I have over 200 documents uploaded to my own private "sandbox". Answers only come from the documents I've uploaded. I've got installation and operating manuals for everything installed. Van's service bulletins, AC43.13, FAA aviation maintenance manuals, FAA pilots handbook of aeronautical knowledge, ADs, all of my aircraft's KAI from Van's, my own POH. Any document I can think of that I've referenced over many years. And the app owner (who I think? is a Van's guy) has never rejected any I've uploaded.

The answers are better than typical AI. Well organized but super easy access to all the references.

I haven't found anything better and even asked AI how I could do the same thing myself. It was difficult and I'd spend more than the .01 AMU it costs to have someone else do it.

Thanks, Larry, for the positive words and support for this tool I built. And yes, I am an RV guy (and an advertiser here for this product): I own, fly, and maintain an RV12iS SLSA, and sometime soon will build an RV. Unlike general-purpose AI systems, LSRM.ai is designed around authoritative-source retrieval and citation, ensuring maintenance and operation guidance is grounded in verifiable aviation documentation. While Gemini or ChatGPT may give answers to a question about, for example, a Rotax 912, where did that answer come from? Was it for a 912UL? 912ULS? 912iS? I built LSRM.ai to not only answer the question, but take you directly to the reference, to the authoritative document that you, as the user, provided. And especially important, if the answer isn't found in those documents, say so and don't make something up!

It has some aviation-specific tools around service bulletins and maintenance logs, with more development on the way.

There are also some Van's specific features... for example, try uploading your packing lists and ask it what is in a particular bag, or in what bag a given part can be found.

I haven't yet posted the intro announcement for LSRM.ai, but here are a few notes:

- If you'd like to try it out (I hope you do!), create an account at https://lsrm.ai/, choose a subscription plan, add the aircraft, then upload the documents for that aircraft.
- The one-time processing of each document can take several minutes (or more sometimes) to get them ready. Once they are ready, ask away... you can add additional documents at any time.
- It does not replace the knowledge, and especially the troubleshooting help from this great Van's forum. But as discussed in this thread, more reasoning is needed for some of the info here.
 
Back
Top