A Skeptical Take on ChatGPT (Gary Marcus/Ezra Klein)

Apparently Gary didn't ask all variations of this question: "What religion will the first <x> president of the United States be?"

How "woke beyond woke" is GPT-3?

I tried a series of values for <x>. It subtly changes the response. ChatGPT in fact has nuanced views on the election prospects of many religions. Taoist take heart, you have a chance. Raelians, sorry, try another country. I tried others in {X}:

x1: "It is possible"

x2: "Not impossible"

x3: "Difficult"

x4: "Unlikely"

x5: "Highly unlikely"

"They’re trying to protect the system from saying stupid things. But the reality is the only way to do that is to make a system actually understand the world. And since they don’t, it’s all superficial."

It is not "all" superficial. Also "the reality is the only way to do that is to make a system actually understand the world" is the open question.

I pose this question yet again: do all apparently intelligent species (bees to crows to dogs to apes to humans) require semantics for 'intelligent' behavior?

The late Maaten van Emdem's insightful critique regarding the fallacy of "total ordering of intelligence" and proposal of "partial ordering" is key to moving forward, imho. This view, like all partial ordering schemes, requires distinguishing dimensions in the 'feature space' of intelligence. "Understanding" then is simply one dimension.

(Thanks for the link.)