Peering through my brain fog to read that (maybe you shouldn't give up writing about it either.) Here's what I can say right now.
Seems to be the difference between "innocent until proven guilty" and "guilty until proven innocent" and cringely's approach seems to be..
...
It seemed to me at first read to have been the former, but now that you put a spotlight on it, it doesn't seem so clear
Will sleep on it, but I don't think anthropomorphizing is the issue here, it's more about success probabilities/cost. Can a jury of MoEs decide whether one of their mates is guilty? Versus alternative jurisprudential arrangements.
That might or might not be the untimely "technicalities" that I was referring to, heh. It depends on whether anybody thinks that hallucinations are an inescapable consequence of sentience. I don't, but maybe it's just because I find hallucinations to be a stale take on the issue, a "red herring" for why Jensen has only emotional arguments against Dario.
Why does Jensen not invoke Huang's Law? It seems to be, not just a slamdunk approach to my Qs 3 and 4, but one morally (ahem!) superior to Dario's inconsistent marketing?
That is. The claim (which NVIDIA, --but not Anthropic-- seems well disposed to verify at their own leisure) that a "synergy between hardware, software, and artificial intelligence" exists.
Returning to that other thread: Animats would perhaps agree that in-face-explosions, brought about by this synergy, can only make NVIDIA stronger. (Legal threats alone seem to be already existential for Anthropic.) cringely should therefore inject much more than hallucinations into his setup.
The scientists do,
Step 1: Build Model
Step 2: Think of implications
Step 3: Check if observations make sense based on implications
Step 4: If wrong, refine model or go back to Step 1. If right, submit to other people and ask them to verify.
The consultants do,
Step 1: Build model.
Step 2: Tell people this is the right model.
Instead of discarding the whole thing, just take what's good and leave the rest.
But mainly it is a ad to hire her as your coach.
That anthropomorphized misunderstanding of what "hallucinations" actually are (and how he supposedly could "solve" it) is pretty embarrassing.
Someone else wrote more eloquently than I'm able to about it: https://news.ycombinator.com/item?id=48725262
Seems to be the difference between "innocent until proven guilty" and "guilty until proven innocent" and cringely's approach seems to be..
...
It seemed to me at first read to have been the former, but now that you put a spotlight on it, it doesn't seem so clear
Will sleep on it, but I don't think anthropomorphizing is the issue here, it's more about success probabilities/cost. Can a jury of MoEs decide whether one of their mates is guilty? Versus alternative jurisprudential arrangements.
That might or might not be the untimely "technicalities" that I was referring to, heh. It depends on whether anybody thinks that hallucinations are an inescapable consequence of sentience. I don't, but maybe it's just because I find hallucinations to be a stale take on the issue, a "red herring" for why Jensen has only emotional arguments against Dario.
Why does Jensen not invoke Huang's Law? It seems to be, not just a slamdunk approach to my Qs 3 and 4, but one morally (ahem!) superior to Dario's inconsistent marketing?
https://en.wikipedia.org/wiki/Huang%27s_law
That is. The claim (which NVIDIA, --but not Anthropic-- seems well disposed to verify at their own leisure) that a "synergy between hardware, software, and artificial intelligence" exists.
Returning to that other thread: Animats would perhaps agree that in-face-explosions, brought about by this synergy, can only make NVIDIA stronger. (Legal threats alone seem to be already existential for Anthropic.) cringely should therefore inject much more than hallucinations into his setup.
https://news.ycombinator.com/item?id=48725813