> My God... this woke thing again...

No, I'd actually say this issue isn't. They can't even make the thing generate what they want even with specific instructions. It's not some fretting that the frequencies don't match modern liberal political sensibilities about "representation," or a hamfisted attempt to over-correct and satisfy those sensibilities.

> or a hamfisted attempt to over-correct

I don’t know anything specific about Meta’s AI, but this could easily turn out to be a case of a hamfisted attempt to over-correct by weighing the Asian part of the prompt so heavily it manages to cancel out the Caucasian part.

Ah yes. Pointing out that AI has biases is "woke." Very good.

Not that part, but the editorializing towards the end gets pretty woke.