Using LLMs has become commonplace with regular coding and game dev, so I thought I’ll try it with a no-code engine too as an experiment.
My experiences (all with free models):
The built in LLM - 0
It completely fails to actually read my event sheets and just lies about what it says. “oh, the error is that this condition checks if this boloean is false, but it should check if its true” it says, while I can see the condition actually does already check if this boolean is true.
Gemini - 4/5
It doesn’t have the most correct answers, but I love its formatting. Its very easy to read. Plus it outputs insanely fast compared to the built in and Claude.
Claude - 4/5
Claude gives the most correct answers, plus it often adds plenty of useful context and suggestions. This one really helps you actually learn what you’re doing and gives ideas for other ways to do it. But I removed one star because the formatting is all over the place, sometimes it tries to format it like Gemini but worse, sometimes it uses its “step-by-step” layout which isn’t very nice to look at. It also needs to load for a long time.
ChatGPT - 2/5
I generally don’t like what this LLM has become and its the same when you ask it how to do things in GDevelop. Its needlessly ego pleasing, constantly apologizing, and keeps getting everything wrong.
Overall I’m not exactly blown away and don’t think I’m gonna be using LLM’s a lot. I think they are good if I’m really stuck on something and just need a fresh idea on how to tackle it. Even if they are only 80-90% correct its still enough to help me figure out the rest.
Have you had any success using an LLM to help you with GDevelop?