The context for this is a top-down movement logic where the character stop is jarring. I couldn’t figure out why it was happening (I’m a novice) so I decide to ask google chrome’s ai. The advice seems sketchy to me. I am wondering 1. does this ai suggestion seem reasonable to a more experienced game designer and 2. what experiences have other people had with asking google’s ai or ai in general.
Please use the Epic AI Assistent which is Epic own way of AI
It depends on your ability to ask questions.I frequently use Gemini and ChatGPT, and both are useful.
My impression is that Gemini is concise, while ChatGPT tends to be quite persistent in providing detailed explanations.
Gemini designed the material function for the 16-bit color mask channel for me, and ChatGPT helped me come up with ideas for a broader game experience.
They remembered my past questions and gave me advice on how to integrate the game experience with the actual design.
Especially for beginners, you’ll frequently encounter unfamiliar features in the answers, which I think will be helpful for learning.
Hey there @dexterbradley! LLMs can be quite useful when you can provide enough context for them to come to the correct conclusions and there’s enough (positive) training data on subject. For Unreal Engine, since there is a decade of training data for multiple versions, there can be some mix ups with version specific information but for the most part the general idea is usually sound even if it misses some of the specifics. To add, since Blueprints are a binary filetype and the main method of working with them is visual, training data for them is somewhat noisy. Hallucinations are still always possible, and still somewhat common at this stage of AI development.
Suffice to say, focusing your questions, having the AI include sources (for you to learn and verify), and mostly utilizing it for things it’s more suited to can go a long way in making their output much better. Prompt work is also a relevant skill here.
Because LLMs don’t actually know anything, they just regurgitate patterns found by mass trawling the internet. Of course you are gonna get sketchy results for any non-trivial problem. 
>they just regurgitate patterns found by mass trawling the internet
In other words, what they’re doing is the same as what humans do. Humans, too, are fundamentally ignorant and need to obtain information from external sources.
It’s common to spend hours searching the internet for information while studying, so being able to reduce that to just a few seconds is of great value.
You don’t see a difference between understanding methodology vs. parroting words that statistically precede or follow each other?
According to that theory, you shouldn’t be using Google search. If you search, you’ll see the same article and the same sentences every time. And search engines will learn nothing other than what kind of ads they should display.
Current AI is a very fast alternative to search engines. It comprehensively analyzes the search results and extracts only the necessary parts. Even if they are parroting words, I think they can be helpful for learning something if you encounter them for the first time.
And of course, you still have the option of learning nothing. We often hear news about AI solving problems that humans couldn’t solve, but you could also ignore that and continue talking about the AI from three years ago that simply parroted everything.