A few months back, I had a great idea of making a game with Chat-GPT and the graphics with DALL-E. this was an experiment to see how useful it was for someone who has no experience with Javascript, but a basic understanding of programming.
I did run this by @TomFulp before hand, to make sure it was all okay to submit to the portal, and I also enlisted the help of @Olskoo to do the Newgrounds API intergration because I couldn't work it out, and I don't think Chat-GPT would have stood a chance.
The game I had a concept for, which I gave to Chat-GPT, and it produced a base structure, and the bones of the game. Over time, I would ask it to fill in parts to add function to the game, such as the generation of planets, the UI elements, and other parts. The only part I played was putting the code into place, like the simple monkey brain I am.
However, when actually getting things to work, Chat-GPT suffered from what could only be described as Computer Senility, fans of Red Dwarf will note that the ships computer, Holly suffers from this affliction. It had a terrible habit of forgetting what the game was doing, what the assigned variables were, and getting stuck in debug loops.
Multiple times, I would have to start from scratch because it just painted itself into a corner, and I couldn't help, because I had no idea what I was doing either, the old saying of the blind leading the blind, nither of us were going to be able to debug the problem.
Another issue I ran into was the character limit for input, so the game came to a conclusion once I couldn't paste the whole code in anymore. I'm not sure if this is a limitation of the free Chat-GPT or not, but there we go. I could have seperated the code into different files, however, again I had no idea what I was doing, and I tried to go down this route with Chat-GPT to do it, however, it was unable, and ended up in a debug loop of hell.
Chat-GPT would also sometimes just stop when giving an output, so you'd say "continue" and it would continue, however, it was 50/50 if it had remembered what code it had written before it stopped, to what was coming next, it was a mess, and frustrating.
The easier part was the art, you just input some words, and get an outcome, though I will say, the mines look more like birthday cakes. I'm not a fan of using AI art for anything outside of personal projects, generally I am more than happy to pay an artist for their time. The words for the description, and in game were also generated by Chat-GPT, and that's probably one of it's better uses, even though it is a pack of lies.
In conclusion, the outcome is a playable game, it sort of matches the vision I had, and for that, it would be a great basis to pass onto a human as a proof of concept to build a game from, with a decent game loop, and have more of a purpose to it. Currently, it is bare bones, an experiment of a monkey brain, and a senile AI.
What did I learn from this experiment though? That AI could be a great tool for what I mentioned above, and for those with an understanding of programming to get it to do some of the more tedious work for you, but you would then be better off debugging it yourself, or getting someone else to look over the code, else you may also end up in debug hell.
You can check out the project below, and please give me your thoughts on this experiment, I am not looking for an ethical discussion, personally I believe the art, and programming would have been better if I had run it all past a human. The biggest ethical issue is shovelware, but that has always been an issue, asset flipping, and dodgy labour.
I would also like to invite @Olskoo to give his thoughts, as he saw the code! I may also release the code to supporters of Newgrounds in future.
Thank you for reading my essay, if there are any teachers, please give me a grade, I hope I pass.
ToasterDude838
So what it seems, as someone who has never used Chat-GPT, is that it kinda suffers from basically short term memory loss but for AI.
Little-Rena
Yep.