You're absolutely right. It's not intelligence in the way we have traditionally understood it. It's a pattern matching system that uses massive scale to handle a vast number of cases, and it's very useful, but it routinely fails on even basic tasks. The labs are just playing whack a mole with the failure cases because fundamentally the architecture doesn't have common sense in the same way we do. Your moniker of "statistical code synthesis" is probably more accurate but this is a capex intensive technology and you're gonna get a lot more investor interest by playing to people's imagination with terms like "intelligence" or "superintelligence".