Chat GPT and MSFS

[Spit40 197]

I’ve been doing a lot of experimenting with ChatGPT recently and getting amazing results. Here’s something i tried just now by way of an example. It was inspired by an issue I had a while back that I thought the tint on the Beaver windscreen was a bit strong. After a number of weeks I eventually found a mod on flightsim.to. It looks like I can get chat GPT to write mods for me now. I’ve actually tried a few AI chat programs, and this particular conversation was with mistral AI. 
 

i said “i find the tint on the Cessna 152 windscreen in Microsoft Flight Simulator a bit too strong. Can you tell me how to put a file in the community folder to override it with something less strong”

Answer

To reduce the windscreen tint for your Cessna 152 in Microsoft Flight Simulator, you can modify the aircraft's material file. Here's how to do it:

1. First, locate the aircraft's material file. The Cessna 152's material file is usually found in the simulator's base package, but we want to override it by creating a copy in the Community folder. Navigate to the Community folder:
`C:\Users\YourUsername\AppData\Local\Packages\Microsoft.FlightSimulator_8wekyb3d8bbwe\LocalCache\Packages\Community`
2. Create a new folder in the Community folder with a unique name for your modification, e.g., "Cessna152\_WindscreenTintAdjustment".
3. Inside the new folder, create another folder named "simobjects" and then create a folder named "Airplanes" and finally a folder named "Asobo\_Cessna\_152" (or the appropriate folder name for your Cessna 152, if it's a third-party aircraft).
4. Locate the original material file for the Cessna 152. It's usually found in the simulator's base package at:
`C:\Program Files\WindowsApps\Microsoft.FlightSimulator_<version>\simobjects\Airplanes\Asobo_Cessna_152\materials`

The material file is named "Asobo\_Cessna\_152.xml" or something similar.

5. Copy the original material file and paste it into the "Asobo\_Cessna\_152" folder you created in the Community folder.
6. Open the copied material file using a text editor like Notepad or Notepad++.
7. Look for the section related to the windscreen tint. It might look something like this:
```xml
<Material>
    <Name>glass_tint</Name>
    <Diffuse>0.2, 0.2, 0.2</Diffuse>
    <Specular>0.5, 0.5, 0.5</Specular>
    <Roughness>0.5</Roughness>
    <Metallic>0.0</Metallic>
    <Emissive>0.0, 0.0, 0.0</Emissive>
    <AlphaTest>0.5</AlphaTest>
    <DoubleSided>true</DoubleSided>
    <Transparent>true</Transparent>
    <BlendMode>ALPHA</BlendMode>
</Material>
```
8. To reduce the windscreen tint, adjust the values in the `<Diffuse>` line. The three values represent the red, green, and blue color channels, respectively. Lowering these values will reduce the tint. For example, you can try changing `<Diffuse>0.2, 0.2, 0.2</Diffuse>` to `<Diffuse>0.1, 0.1, 0.1</Diffuse>`.
9. Save the changes to the material file and close the text editor.
10. Launch Microsoft Flight Simulator and load your Cessna 152 to see if the windscreen tint has been adjusted to your liking.

By placing your custom modification in the Community folder, you can easily share it with others without modifying the default aircraft files. Simply compress your "Cessna152\_WindscreenTintAdjustment" folder into a .zip file and share it on platforms like Flightsim.to or the MSFS forums. Make sure to provide clear instructions on how to install and use your modification.

[Spit40 197]

OK, I'm getting ambitious now. This isn't concluded yet, but if it works my mind is well and truly blown!! I'm sharing WIP just to inspire others. It knows I'm talking about MSFS as I have a chat about that already underway.

is it possible to make a simple mod that will give me just the extra option of storing more than 3 recently used airfields and can you give me some code to do it

Yes, it is possible to create a simple mod that will allow you to store more than 3 recently used airfields in Microsoft Flight Simulator (MSFS). Here is some sample code that you can use as a starting point:

1. First, you'll need to create a new XML file in the MSFS community folder for your mod. Let's call it `recent_airfields.xml`.
2. In `recent_airfields.xml`, you'll need to define a new UI element that will display the list of recently used airfields. Here is some sample code that you can use:
```
<UIElement type="ListBox" name="RecentAirfieldsListBox">
  <Property name="Visible" value="True" />
  <Property name="Width" value="200" />
  <Property name="Height" value="200" />
  <Property name="ItemsSource" value="RecentAirfields" />
  <Property name="SelectedItem" value="{Binding SelectedAirfield, Mode=TwoWay}" />
  <UIElement type="TextBlock" name="RecentAirfieldsLabel">
    <Property name="Text" value="Recent Airfields:" />
    <Property name="FontSize" value="18" />
    <Property name="Margin" value="0,0,0,10" />
  </UIElement>
</UIElement>
```
This code defines a new `ListBox` element that will display the list of recently used airfields. The `ItemsSource` property specifies that the list should be populated with the `RecentAirfields` data source, which we'll define next. The `SelectedItem` property binds the selected item in the list to a property called `SelectedAirfield`, which we'll also define later.

3. Next, you'll need to define the `RecentAirfields` data source in your mod's code. Here is some sample code that you can use:
```
private ObservableCollection<string> recentAirfields = new ObservableCollection<string>();

public ObservableCollection<string> RecentAirfields
{
  get { return recentAirfields; }
  set { recentAirfields = value; OnPropertyChanged(); }
}
```
This code defines a new `ObservableCollection` called `recentAirfields` that will store the list of recently used airfields. The `RecentAirfields` property exposes this collection to the UI.

4. Finally, you'll need to add some code to your mod that will update the `recentAirfields` collection whenever the user selects a new airfield. Here is some sample code that you can use:
```
private string selectedAirfield;

public string SelectedAirfield
{
  get { return selectedAirfield; }
  set
  {
    selectedAirfield = value;
    OnPropertyChanged();

    // Add the selected airfield to the list of recent airfields
    if (!recentAirfields.Contains(selectedAirfield))
    {
      recentAirfields.Insert(0, selectedAirfield);
    }

    // Remove the airfield from the list if it appears more than once
    while (recentAirfields.Count(a => a == selectedAirfield) > 1)
    {
      recentAirfields.Remove(selectedAirfield);
    }

    // Trim the list to the desired length (e.g., 10 items)
    while (recentAirfields.Count > 10)
    {
      recentAirfields.RemoveAt(recentAirfields.Count - 1);
    }
  }
}
```
This code defines a new `SelectedAirfield` property that will store the currently selected airfield. The `set` method for this property updates the `recentAirfields` collection whenever the selected airfield changes. It adds the selected airfield to the beginning of the list, removes any duplicates, and trims the list to the desired length (e.g., 10 items).

[mspencer 432]

And does any of this work?

Generative AI is a fundamentally derivative product. When GPT3 was released to the masses a bit over a year ago, people posted here having it explain all sorts of things to them. Unfortunately, in most cases, it was completely hallucinating and the results were meaningless.

A lot of attempts to have it synthesize code are either... shockingly basic (comp sci 101) or completely incorrect.

Edited March 3 by mspencer

[Spit40 197]

9 minutes ago, mspencer said:

And does any of this work?

Generative AI is a fundamentally derivative product. When GPT3 was released to the masses a bit over a year ago, people posted here having it explain all sorts of things to them. Unfortunately, in most cases, it was completely hallucinating and the results were meaningless.

A lot of attempts to have it synthesize code are either... shockingly basic (comp sci 101) or completely incorrect.

Don't know yet, but my experience of getting it to write Add In scripts for Fusion 360 CAD software was that it needed coaxing. As long as I was able to get errors of what wasn't working and feeding them back it kept "apologising" and giving me suggestions until it worked. At one point its training was too old for the latest API, so I found a sample of code from the new API and passed it into the chat. It said "Thanks very much that helps me ,,," and we got it working. All this is in a language I have no experience of. That was pretty complex stuff, I suspect the window tint or other simple config file things will work.

I should add that I tried some MSFS stuff a month or two back in Chat GPT 3.5 and wasn't impressed. I'm not paying for Chat GPT 4 but I read that Mistral AI is almost as good as 4, so I was doing this in Mistral AI.

Edited March 3 by Spit40

[History 42]

Are you using the paid version

[Spit40 197]

I'm using https://chat.mistral.ai/ which is free and claimed to be almost as powerful as the paid version. 

[superspud 450]

I was under the impression it stopped 'learning' in 2021 which means a lot of MSFS has changed. 

I tried it with a couple of things relating to it that were tormenting me but it spat out generic and useless answers. 

Edited March 3 by superspud

[virtuali 4,421]

Don't post your impressions about ChatGPT if you haven't tried the paid 4.0 version first. It's night and day difference compared to 3.5 and it's also much better than the free one included in Bing and the cutoff date for its knowledge base is December 2023 right now.

[RobJC 2,071]

That is awesome. 

[cianpars 1,171]

Someone posted here a while back that ChatGPT could produce flightplans for you.  I tried it and it did come up with what looked like a fairly reasonable flightplan.

This technology will of course get better in time as it is always learning 

I would imagine that it could do a host of flightsim related things for you but whether the output would actually work, I'm not so sure.

Try asking it to find an addon airport in an area you are interested in or a piece of scenery.

Edited March 3 by cianpars

[leprechaunlive 1,859]

Dont bother wasting time on the first mod, the whole set of instructions is 100% WRONG, about basically everything (method, file name, etc).

Its the trick with ChatGPT, sometimes it "thinks" it knows, and spill out answer thats look very convincing, but actually, its basically "adapting" a commonly asked/known solution for similar use cases to other application where it doesnt necesserily apply. 

 

MSFS dev envirronment is very, VERY, specific, quirky, (some would say outdated lol), so ChatGPT is usually useless for it.

 

PS: quickly looked at your second "mod case", same as the first, dont bother 🙂

Edited March 3 by leprechaunlive

[Lorby_SI 2,042]

I made a few experiments integrating ChatGPT into the "simulator workflow", for example as a METAR reader or flight attendant. Even the older 3.5 works OK for that.

If you are interested (and are a user of the AxisAndOhs app), here are some examples:
AAO 4.10: having a chat with ChatGPT - AxisAndOhs Support - The AVSIM Community

What I found really astonishing are the AI services for voice recognition and text-to-speech. For many years I was used to what Windows had to offer. Even with good SAPI voices (I got a few from Cereproc), it was always obvious that a robot was talking to me. This was especially noticable when I made one of the voices say sentences in a different language. But now, be it with Amazon Polly or Azure Cognitive Speech, the results are nothing short of amazing. (IMHO of course).

Edited March 3 by Lorby_SI

[mspencer 432]

3 hours ago, leprechaunlive said:

Its the trick with ChatGPT, sometimes it "thinks" it knows, and spill out answer thats look very convincing, but actually, its basically "adapting" a commonly asked/known solution for similar use cases to other application where it doesnt necesserily apply. 

Shoddy use of terminology has resulted in millions of people with access to the internet treating LLMs like AGI. 

The LLM isn't actually doing any thinking - it's stringing together associations and doing its best to put together something digestible, which it does very well, but it basically doesn't know anything about what it's telling you - it just scored the associations well.

As a result, LLMs are great to do things like... write content like LinkedIn posts, answer basic questions, and communicate non-complex information to a variety of audiences. Anything you're very familiar with or an SME in, if you ask any of the LLMs even perfunctory questions, they do a great job of projecting confidence but usually get key facts wrong or start completely making up new facts. It's because the model can't say "I don't know" and it's not actually doing any reasoning.

For example, sticking with the theme, here's a great use of an LLM, if you wanted to say, create a VA. It can do creative things just fine, but just know all of its work is derivative:

[bobcat999 8,489]

5 hours ago, leprechaunlive said:

Don't bother wasting time on the first mod, the whole set of instructions is 100% WRONG, about basically everything (method, file name, etc).

Its the trick with ChatGPT, sometimes it "thinks" it knows, and spill out answer that's look very convincing, but actually, its basically "adapting" a commonly asked/known solution for similar use cases to other application where it doesn't necessarily apply. 

....

PS: quickly looked at your second "mod case", same as the first, don't bother 🙂

The 'Chaun is back!   

How are you doing chap?  Working on anything yourself or with Big Radials?

I loved the Norseman.  I need to get it back out of the hangar and get some air under it's wings again. 
Things move so fast.  I am buying more planes than I have time to fly at the moment.  Sadly there is some good stuff that I worry will never see enough hours.

PS. Totally agree with your post above by the way.  But it just shows it is learning really well from us humans, as there are a lot of BS-ers out there in real life as well.  

Edited March 3 by bobcat999

[David Mills 4,774]

I have no experience with the paid version of Chat GPT. But the first free iterations I fiddled with provided boatloads of misinformation. And I'm not talking here about anything related to liberal-versus-conservative philosophy. I'm talking about completely empirical facts such as the location of famous individuals' gravesites. Chat GPT was especially terrible in this regard. I would repeatedly ask it where Neil Armstrong was buried. It provided vacillating answers such as Wapakoneta, Ohio (his hometown), Columbus, Ohio, Arlington National Cemetery, etc. Often Chat GPT provided highly specific details about the cemetery environment itself and the exact wording of Armstrong's headstone. In reality, however, Armstrong was buried at sea and has no headstone and no gravesite at all. I could name two dozen other such examples involving gravesites, from the Wright Brothers, who, according to Chat GPT's initial release, were buried in Cleveland, Ohio, to country singer Hank Williams, who was supposedly buried in Nashville. Both of these answers are wrong. Chat GPT often amazes me when it provides an incredibly accurate and highly useful answer. But it is also the most convincing liar of all time. I'm just happy I didn't plan a visit to Armstrong's grave based on its answers.