I use it for coding, with mixed results, for getting me going in the right direction. I take everything it says with a grain of salt. Through some original queries about some people, I came to realize that a GPT is a stochastic parrot optimized for plausibility. Given a Venn diagram of plausible and reality, they have a large overlap. Our ideas of plausibility are informed by common repetitive observations of reality…
The library/language versioning problem is going to be a tough one to overcome, because most of the training data available doesn't specify the version of either, so an AI's response will be in an unpredictable version or even mix of versions.