From the few days of testing I can clearly say that it only works if you know how to code, meaning that for beginners, this isn’t going to help… Much, at least.
As mentioned on this post, albeit recommending it to new scripters, it’s of no use because the code that the AI generates is not perfect. Additionally, the AI only writes based on what exists, so if you’re not careful, it’ll write on its own mistakes. This meaning, if you don’t have a clear knowledge of the language, what use will a new scripter give to it? They’ll expect it to be perfect, or simply won’t see mistakes.
Another problem is that the AI is extremely slow, at least compared to Copilot. For a single line of code, Copilot suggested it in <0.7 seconds, while Generative took around 3 for the same result. Is this the cooldown between line pause and suggestions, or does it take time to compute solutions?
Since GenerativeAI uses free model code, the outcome is also generally not the best. It cannot typecheck, uses deprecated globals (not looking at wait
), and suggests beyond what we can call “good” code. The current version is plagued with bugs and does not promote using more performant and recommended functions (like task
). It does, but not out of the box, so another no-no for what you want to achieve as a solution for beginners.
It’s a beta version, yes, it is unstable and has its issues. However, this AI really needs to be trained using a larger data collection. Free model code, as everyone knows, is not the best in the world. Say what you want, though.
Honestly speaking, you can simply enable a feature in your games that allow data (script) collection for the AI to train on. Much better than using free models. At least then we know our data is being sold.