OpenBOR development with Github CoPilot?

[Digital Brilliance Hour]

Hey guys,

Sorry that I've been gone for so long, we had a LOT of family personal matters that needed to be tended to for honestly about a year and a half, and we're almost back at a more stable situation.

But now that things are getting to a stable point, I wanted to ask how beneficial it would be for you guys if I did series on leveraging Github Copilot when generating code for OpenBOR Games?

I'll still stress that you at least have fundamental knowledge of how the engine works before trying to use GenAI as it's not for people who can't decipher what the AI is doing. I'm also thinking about maybe creating a copilot agent that specializes in solely focusing on OpenBOR projects.

What do you guys think? Would that be useful to you guys?

 

[DCurrent]

Hey guys,

Sorry that I've been gone for so long, we had a LOT of family personal matters that needed to be tended to for honestly about a year and a half, and we're almost back at a more stable situation.

But now that things are getting to a stable point, I wanted to ask how beneficial it would be for you guys if I did series on leveraging Github Copilot when generating code for OpenBOR Games?

I'll still stress that you at least have fundamental knowledge of how the engine works before trying to use GenAI as it's not for people who can't decipher what the AI is doing. I'm also thinking about maybe creating a copilot agent that specializes in solely focusing on OpenBOR projects.

What do you guys think? Would that be useful to you guys?

That would be a HUGE no. LLMs know exactly jack s*** about OpenBOR development because they are trained on public knowledge, which is so objectively wrong I don't even know where to start. Half the world thinks OpenBOR is an emulator, and that's where models get their info.

Worse than that, they appear to know plenty on the surface because that's what they're designed to do. If you are not already an advanced developer, LLMs are going to cost you severely, then cost us in both public perception and trying to support the stupid mistakes they make.

This isn't my being anti AI or talking out of my a**. I'm not only the lead developer of the engine, but also a sitting committee member at The University of Kentucky's Center for Applied Artificial Intelligence. I've set up a small model with specific rules for my own use - it still only gets things correct about 60-75% of the time and tries to tell me "facts" that are outright wrong about the engine or how it works. I can't tell you what to do, but I have to ask you respectfully not to engage in this, and if you do, I will not support you in any way.

DC

 

[Digital Brilliance Hour]

Exactly what I needed to hear.

You had me at "HUGE no" haha

That's why I ask first, so I'm not wasting my time for sure.

 

[Kratus]

Some time ago I tested OpenBOR scripts using Grok, just for curiosity. It even coded the logic correctly, but with a lot of problems mainly in the optimization aspect.
In addition, it completely ignored many script features like localvar/entityvar/globalvar, treating everything as pure C.

 

[DCurrent]

In addition, it completely ignored many script features like localvar/entityvar/globalvar, treating everything as pure C.

That's one of the biggest problems. All the models I tested also want to "correct" you for use of types (ex: void to return pointer, or use of typeof() to enforce type because the LLM doesn't understand the weak typing, and so on).

DC