In terms of human development, publicly available and disclosed machine intelligence has reached what in human child development would be considered a major new milestone.

Previously, Eliza and the likes could only mimic basic speech construction, with increasing levels of correctness.

Now, GPT-3 can mimic meaning and understading, rather convincingly. Its mistakes are akin to a child's naive questions.

I would put its equivalent human age at about 3-5 years old. Sometimes 5-year-olds can come up with nonsense, and sometimes you have an "out of the mouths of babes" moment.

That's really huge, IMO.