Hmmm...

Really, if you want to understand something, look at its simplest cases. If you want to understand failings of ChatGPT 4, know the failings of ChatGPT 3. It's certainly pretty amazing, but it is also certainly limited. I asked it question about CA scuba diving, something I'm an expert on. It wasn't even in the ballpark. I've asked it programming problems, but more, I've watched videos of others ask about programming and it was amazing, but clearly to verify and fix the solutions it generates, you need an expert knowledge.

If it can make the type of mistakes it does, then it is not thinking like human general intelligence.

Something else to think about. I study and have written about human cognition. Logic is a learned skill. It is not our primary problem solving method. Now that's not necessarily a ding, because it is not the primary problem solving method of current AI models either. Both are neural nets and so work by pattern recognition. Human neural nets can spin off logical processes and according to this article, so can AIs. (I also talk about probability boxes humans use. I wonder if AIs can use them as well.)

I have more thoughts on the matter, but it gets long winded. We will see. AIs are definitely amazing tools.