The basic idea is, similar to the "blockout method" in level design, to first generate some raw executable code in GDScript and then,
if necessary, convert it into optimized and parallelized GDExtension C++ code.

The question now would be, if there are any tools or methods that could help with or support this process?
The following approaches seem to be interesting in this context:

https://github.com/godotengine/godot/issues/11068
https://github.com/minosvasilias/godot-copilot

So basically it's about two approaches:

  1. manually optimize the raw GDScript code and then automatically convert it into C/C++ code.

Here the question would be, which usable tools are there, which can convert GDScript code into C/C++ (GDExtension) code?

  1. convert GDScript code into (optimized, parallelized) C/C++ code automatically with an AI tool.

Since I don't have access to Chat-GPT and don't have the time to do extensive tests at the moment,
it would be interesting to know if someone has already made practical experiences with these AI asisstants
(especially concerning the development for Godot) and if the idea that such a tool can convert GDScript code into optimized
C/C++ code is realistic at all. Could this save time or would the effort to manually correct such an automatically generated code
be more time consuming than opimizing and converting it yourself? Due to technical limitations, probably only relatively short code segments could be converted anyway. Nevertheless, it would be interesting to test whether, for example, if specially trained LLMs could achieve some useful results.

Chat-GPT will proudly help you convert your script into C++. But be prepared to adjust his code a LOT. In my experience, the GPT models codes nicely, but they're very stubborn.

What I mean by that is that GPT sometimes creates a magical code with methods and functions that do not even exist in Godot or any other coding language.

What results in you correcting and adjusting the code in order to make it work. ( which opens the question of whether the performance of the script is good too )

But GPT does help to get you on the right track tho.

I don't rely on GPT to convert a whole script once or twice.
But I do ask it for "small" help or overall guidance.

In principle, this also largely coincides with my previous experiences with these AI tools.

The question would be whether specially trained language models could be used for reliable everyday code optimization or conversation.
But it could become problematic to find enough GDScript examples for the training. Added to this would be the costs, time and hardware requirements for the training.