|
|
|
|
|
|
by 7777777phil
124 days ago
|
|
|
well.. exoskeleton vs intern framing matters less than whose cognition gets amplified. AI converges to the mean through next-token prediction, so it amplifies average thinking well but struggles at the tails. The research on this is pretty clear, human-AI pairs only outperform when the human brings real domain knowledge: https://philippdubach.com/posts/the-impossible-backhand/ |
|
|
I have some solidarity with the author's conclusions and doubts in this article, but he substantiates them with very questionable arguments. That is, he presents not some research, but simply one example where nuances are very important.
HOW did they ask about the plausibility of the tennis video, such that the tennis player said it's not people? Here there's a very big difference - either he would simply be watching tennis and say this, or they asked him "Is this a real game or did artificial intelligence draw it?" This is not some significant research (antiplacebo).
And about the architecture of artificial intelligence systems, the author judges by some very popular-science simplified models. But real models are of course unknown to the general public and are corporate secrets.
The fact that one tennis player pointed to the unrealism of the video doesn't mean he would be able to explain in an artificial intelligence system how to make a realistic video.