← Back to context

Comment by biophysboy

9 months ago

I like chollet's definition: something that can quickly learn any skill without any innate prior knowledge or training.

That seems to rule out most humans. I still can’t cook despite being in the kitchen for thousands of hours.

  • Then you're not intelligent at cooking (haha!). Maybe my definition is better for "superintelligent" since it seems to imply boundless competence. I think humans are intelligent in that we can rapidly learn a surprising number of things (talk, walk, arithmetic)

    • > I think humans are intelligent in that we can rapidly learn a surprising number of things (talk, walk, arithmetic)

      Rapid is relative, I suppose. On average, it takes tens of thousands of hours before the human is able to walk in a primitive way and even longer to gain competence. That is an excruciatingly long time compared to, say, a bovine calf, which can start walking within minutes after birth.

      1 reply →

I like Chollet's line of thinking.

Yet, if you take "any" literally, the answer is simple - there will never be one. Not even for practical reasons, but closer to why there isn't "a set of all sets".

Picking a sensible benchmark is the hard part.

  • I think its more of a measurable quantity than a intelligent/non-intelligent threshold binary. Chollet literally made a paper defining it as something like (skill*generalization)/(experience+priors). I don't think its a flawless model, but also I didn't create keras