Website powered by

My *initial* thoughts on ChatGPT

Article / 06 December 2022

Hello friends,

I know it's been a while since any activity in here. Fret not, I'm doing good and I hope you all are too in these crazy times. This last year has been one of self-improvement & exploration, which meant that art sadly took a back seat in many occasions. I'm slowly figuring things out and I'll be back on personal art soon enough! Some of you might have noticed that I deleted my Twitter account, in reaction to the unethical way the new management has handled things and its employees. Thus from now on I'll be focusing my attention on more frequent blogposts here and maybe on Linkedin. Feel free to connect with me there as well if you want to keep in touch.


In this blog post I want to note down a couple of my initial thoughts on the latest AI thingy called ChatGPT. The TL DR version is that it is both amazing and scary to say the least. This is not yet another image generation tool that will replace our jobs tomorrow but still, I cannot help but wonder at its potential to change a much wider range of industries for better or worse. Impactful is certainly the word I'd use to describe it. Of course it's still early days and I haven't had the chance to mess around with it for more than a day or two, so please take everything I mention here with a huge grain of salt! Do your own research and try the tool for yourself, to get a better understanding of how it works and what its (current) limitations are. Also keep in mind that I'm not promoting the tool in any kind of way, I'm simply exploring how we can utilise the tool as artists to enhance our craft.

Initial tests

The tool came to my attention after some friends mentioned that it is apparently good at writing code. In order to properly test it, I head out to create a simple Maya script. After a solid 5 minutes and with no previous knowledge of MEL, I managed to get something working. Mind you this was a super simple task and I I couldn't do the same for Blender in that specific case (was getting errors, probably due to API changes in the last year or so). Still, I feel like the screenshot below showcases the tool's power. It can quickly generate code, comment it, explain what everything does and how it works and I can definitely see it as a great educational tool by that fact alone!

What's even crazier is that you can continue the conversation and ask it to make updates to the existing code, inform it on errors that your script editor might throw and give it feedback on how improve the code. I might not be the best person to judge it in its code efficiency, style or if some things work or not, but the fact that it even is to that level already is both scary and revolutionary in my eyes. I've tested "AI" chatbots in the past but they have always been fairly easy to confuse and quickly end up with a mess of a result, so this really looks like a proper AI companion that actually understands what you ask it to do and has the ability to iterate on its answer. I cannot wonder but feel like we're getting very close to an actual Tony Stark - Jarvis moment, I certainly feel like that every time I chat with the AI. Imagine how better it will be in 1, 2, 5-years' time!

Stress-testing it

While playing around with simple Maya scripts, I wondered if it could write shaders for Unity, Unreal etc. Turns out it actually can! I tried some Shadertoy code with a friend and that worked surprisingly well. Next up we tried a simple Unity shader and even asked it to write a custom Kuwahara post process filter in Unreal. That turned out to give me lots of errors and I'm not tech-y enough to fix them but the base algorythm code looked correct, so I'm sure any Tech artist would actually be able to set it up properly and get it working.

What I want to stress out is that this tool isn't panacea, you still need to be knowledgeable in your area of interest and adjust what the AI gives you to fit it to your needs. It still feels like it missed the last 10% of polish and accuracy in its answers for me to call it perfect. Nevertheless I feel like, it can already be a huge help to anyone. Be careful though! The tool will be 100% confident that whatever it tells you is the absolute truth and certainly right. In many situations the AI might give you complete bs but it'll still be absolutely certain that it gave you the correct answer! That's where your skillset and proficiency comes to play, to filter out the good from the bad answers.

AI's role as a brainstorming partner?

While I was looking out of my train's window the other day, I stumbled across an abandoned kart circuit and thought that this would be a nice environment idea that hasn't really been explored much in the env art community. Just to test the AI out a bit more, I gave that idea of mine a go, asking the AI for potential theme/moods and what props I should include. I encourage you to read its answers below and let you be the judge of that...

Throughout the whole conversation the AI gave me excellent suggestions that helped speed up the initial environment research process and gave me lots of ideas that I might not necessarily have thought. Some of the answers it gave me were pretty generic and predictable of course (who would have thought an overgrown & abandonded place would have overgrown vegetation and abandoned human props, right?), but I'm sure I could ask it to pump out extra ideas and it felt like there was definitely room to push it further! I cannot wait to continue the thread and see how far I can go with that brainstorming mindset.

Final thoughts

The reason of writing this blog post was to mainly spark a conversation on how we can use this tool in our creative process. We need to be part of the conversation as artists, understand how these tools are developed and provide both feedback and critique to their creators. I'm not aware how exactly this specific AI model has sourced its data and I will most definitely not be condoning its use if I later find out that OpenAI used similar unethical training data collection techniques as Midjourney & Stable Diffusion. You can read more on OpenAI's latest blog post. If you've got any extra resources on the subject, please do so in a comment. I'd love to hear more inputs on this subject!

Till next time,
*flies away*