It's not because they can't write PRs indistinguishable from humans, or can't write code without Emojis. It's because they don't want to freak out the general public so they have essentially poisoned the models to stave off regulation a little bit longer.
That's a lot of expensive work they're doing, and ignoring, if they're just later poisoning the models!
I'm like "Sure buddy, sure. And the nanobots are in all vaccines, right?"
Of course not! They would use it to trade and would keep it concealed while throwing the public a bone with a less advanced version. Same thing applies as AGI or even as code gen gets better.
How likely is it that you saw a plane made by humans that you misclassified due to lack of data vs how likely is it that it was actually an Alien made aircraft, coincidentally working in our atmospheric conditions, gravity and drag related physics?
Guess what, it's about likeliness and you are extrapolating the wrong assumptions. Extraordinary claims require extraordinary evidence, not the other way around.