Raw LLM Responses

I think you just don’t know how AI works. Yes, to an extent you are right they aren’t looking for perfect spelling. You are however, reaching the wrong conclusion. If you give ChatGPT instructions on how to write that are more specific like avoiding specific words or varying sentence length, then it is going to be far less conclusive if you’ve used ChatGPT or not. There are a couple of terms that are very AI coded like delve into and tapestry. AI detectors can’t really know if an AI wrote something unless open AI decides to implement a detection system where they imperceptibly change text generated to be detectable. The thing is that someone who is autistic, speaks English as a second language or has just worked through their text a lot it’s going to have fairly similar output to ChatGPT. It’s not as much detecting the colour of the pencil as detecting if the pencil is red or not. If you ask it to write in different ways, it’s going to be a pink where the average person would be a white pencil. It’s going to confuse the results enough that it isn’t necessarily going to be detected as AI. It’s especially hard to detect formal writing which is already very rule following, not to mention how most educational institutions allow stuff like Grammarly which also makes a lot of of texts look AI generated.