On AI

My policy on AI is: use and abuse, but not all the time and not at all costs.

As a solo developer, it is my responsibility to use every tool at my disposal to bring my game to fruition. But that also means knowing when not to use something, especially when it ends up slowing me down or when the results are subpar.

ChatGPT has been a great advisor on both the creative and technical side. I’ve used it to troubleshoot technical problems, brainstorm ideas, organize systems, and give structure to thoughts that would have been harder to untangle on my own.

Simply taking a screenshot of a broken Blueprint and asking, “Why doesn’t this work?” has saved me countless hours of debugging, only to realize I plugged in the wrong node or made some silly mistake. The chat immediately notices these kinds of things and alerts me to errors that I might have stared at for another two hours.

More importantly, it helps me learn and use things quickly that are way outside my expertise.

Case in point: math.

All those formulas I don’t remember from high school, and all the ones I never learned in the first place? The AI knows them. More importantly, it can guide me through implementing them in Blueprint, C++, or any other language I ask it to use.

The other day, I asked it to help me implement a change in volume over time based on a graph.

Done.

In addition to technical issues, it has also been a great help to me in writing. All those spelling and grammar mistakes I used to miss? It finds and fixes them. And if I ask why, it gives me an in-depth lecture on the English language.

If I get stuck on a piece of writing, I sometimes ask it to simply continue it.

Ninety percent of the time, what it comes up with is bad in one way or another. But seeing where I don’t want to go often helps point me in the right direction. And once in a while, it comes up with a truly brilliant tidbit.

What I always keep in mind is that it might be wrong.

The problem is that when it is wrong, it is usually one hundred percent convinced that it is right. That can lead you down a two-hour rabbit hole in completely the wrong direction. This can happen with technical issues, research, writing advice, or almost anything else.

So I always try to stay alert for anything it says that does not quite make sense.

Ultimately, it is on me as the user to pay attention.

On gen-art, music, and voice acting, I’ve used ElevenLabs for prototyping “voice acting” and Suno/Udio for music. The results are okay. It is fine for prototyping, and it is very helpful to walk through the level and hear everything together.

But I’m very undecided about the final product.

Especially with voice acting. As it stands, AI still sounds robotic and stiff, or over-the-top, depending on the specific voice. From my experience, real voice actors still have the definite advantage in quality.

The problem is that it might not be a choice between AI and voice actors, but between AI and nothing.

I will give a more detailed approach on how I work with those tools in a dedicated post.

As for 2D and 3D art, I’ve found it easier in most cases to use asset packs that are available for purchase in the Fab store, then mix them together to get the environment I’m looking for. The main reason is that it is quicker and easier to get modular assets that way, without having to invest the time and money required to generate each and every asset.

On occasion, though, I find myself needing a specific unique asset that is unavailable on the store. In that case, I might generate it with AI, especially when it comes to a temporary asset I would need to replace anyway.

This is how I’ve been using AI on my project so far.

If you have advice, ideas, or any random thoughts on the subject, please add a comment below.

See you again next time.