chatGPT has trouble with construct 3

My test of the AI still comes up short on construct 3. I ask for the count of sprites overlapping another sprite. They all say use sprite.count instead of sprite.pickedcount. If you question the answer they redo it and explain it correctly, but still. Their first try is wrong. When I get an AI that gets this question right I will let you know.

You do realize AI queries don't actually do any thinking or reasoning for you right? They basically just aggregating whatever relevant responses they can find that sound about right and dress them up to be presented as an answer.

Just makes it harder for people who have no idea to recognize their search result is mostly nonsense but they can't tell because it sounds kinda right.

I find it extra hilarious when some helpful person offers an AI answer to someone else's question, when they aren't familiar with the topic enough to even realize that the help they are regurgitating doesn't make any sense at all in the first place.

I find chatGPT quite useful for scripting. While it doesn't know much about the Construct API and often makes up things that don't exist, if you give it the right information, it can produce working code.

You do need to use right types of AI, ChatGPT, copilot, gemeny are useless

Our models are Claude, and HTML5 code on poe.com

They knows Construct 3 well, but you still need to ask them right

Do these get the sprite overlap question right the first time?

It's useful for general scripting questions, but Construct-specific stuff it gets wrong.

you can teach the AI a little, for example, ask to not use JS things when doing formulas for events. Inside one conversation, the AI will follow the context and give you the right answers