The semantic connections you are talking about are connections between sensory inputs and pre-existing structure inside our brains. You're just reducing what it's doing to the bare basics of its mechanics, but its impressive behavior comes about because of how massively complex the structure is.
I don't know if you've tried it out, but it doesn't just "get lucky". Imagine a student passing one test after another, would you take someone telling you they only "got lucky" seriously, and if yes, how many tests would it take? Plus, it can successfully apply itself to problems it never directly encountered before. Yes, not reliably, but enough that it's beyond "getting lucky".
You talk about it like you haven't actually tried it out. It's not at all the same as previous chatbots, it has really impressive capabilities. It can give you correct answers to unambiguous questions that are non-trivial and that it has not specifically encountered before in its training. And it can do that a lot, repeatably. Nothing to do with how confident it sounds, I am talking about unambiguously correct answers.
Again, I'm not saying it is reliable, but you are seriously downplaying its capabilities if you think that's all it does and I encourage you to try it out for yourself. Especially when it comes to programming, it is incredible. You can put in it complicated code that is undocumented, and it can explain to you what the code does exactly, what problem it probably was intended for, and how to improve it, and it works a lot of the time, much more frequently than "luck".
If all you want to say is that it isn't right all the time, then yeah, that's true. It's very, very frequently wrong. But that has little to do with what you are describing. It could (and will) improve significantly on accuracy, using the same mechanism. And practically, what you are saying doesn't matter. A database doesn't "know" what something is either in your sense of the word, neither does a web crawler, or anything like that. That doesn't make them unreliable. Neither is a human reliable because they "know" something (again going by your definition).
ChatGPT is unreliable because we observe it to be unreliable. That requires no explanation. What does require explanation is why, as
@Demystifier said, it is so much more reliable (especially at non trivial, "reasoning" type problems) than you would naively expect.