Truth be told, it doesn’t really matter how hard the problems are. NN can learn and resample anything in their training data distribution. It’s really a challenge to distinguish between learned and resampled knowledge vs. creative intelligence. So far, nothing in terms of creative intelligence has been demonstrated, AFAIK.
Later, if we get there, it’ll be hard for human experts to assert if AI model-generated new knowledge via creative intelligence is valid or not.
But not better for creators who are still being ripped off. In fact worse, as they attack copyright ©️ altogether!!
Truth be told, it doesn’t really matter how hard the problems are. NN can learn and resample anything in their training data distribution. It’s really a challenge to distinguish between learned and resampled knowledge vs. creative intelligence. So far, nothing in terms of creative intelligence has been demonstrated, AFAIK.
Later, if we get there, it’ll be hard for human experts to assert if AI model-generated new knowledge via creative intelligence is valid or not.
"So far, nothing in terms of creative intelligence has been demonstrated, AFAIK."
My impression is that that's no longer correct. But I will defer to experts.
$5 says o3 will be better, not worse, than other models at most common-sense tasks!