It's still statistical analysis of letter patterns. Just analysis on such a huge scale. In my mind, it's a solution without a problem. MS word has been using pattern analysis to find misspelled words since the 90's. Didn't need to grab huge amounts of data to do it.These AIs are a bit different. They use pattern recognition. So for example if of all of the data they saw the most common letter is "a" then they would only say "aaaaaaaaaaaaa" and nothing else (not to mention feedback into themselfs to further want more "a"s) but what these do is if they see they "appl" was just written they would know from probability that "e" was next. That is also how they answer questions. This is a much newer idea.
For your initial question regarding circuit design and analysis, it wouldn't be an inappropriate use of AI to have it digest solely schematics and circuit diagrams, especially if you could vet the data going in for correctness. Then ask it to design a circuit. This is a focused task and a more appropriate use of Machine Learning.
Today's LLMs have digested schematics PLUS a huge amount of other data. There's too much noise to arrive at a reasonable statistical result. Additionally, the effort and energy used by an LLM is an absurd waste.