Less sarcastically, ∞ as an extension to the number line which has the properties n / 0 = ∞ and ∞ > n for all n != ∞ is both in common parlance and can be rigorously formalized.
Sounds like they're saying that since the distribution doesn't come from measuring or calculating the probability of something, it has the form of a probability distribution but isn't really one. Like saying 5 feet is a height that a person can have, but since I just made up that number it's not actually a person's height.
The soft max is the probability of the next token being whatever in the training data conditioned on the inputs. The author just doesn't know that apparently and thinks it was an arbitrary choice.
The author's essay on the sigmoid similarly lacks the deep understanding that it comes from somewhere and isn't an arbitrary choice.
Softmax isn't a loss function. It is used to transform model outputs into positive numbers that sum to 1, so that they can be interpreted as probabilities, and then those numbers are passed into (typically) the cross entropy loss function. I think you mean, which models are trained using some function other than softmax to transform the model outputs. There are a number of alternatives to softmax, such as the ones described here https://www.emergentmind.com/topics/sparsemax
They’re not. Cross entropy loss is E[-log q] where q is a probability. You could convert the model outputs x into probabilities using some other function like q = 1/Z x^2, and compute cross entropy loss just fine.
iirc, there is a bunch of formal
machinery you need to define probability distributions for situations such as infinite outcomes (eg what is the probability that a random real number between 0 and 10 is less than 3?)
Close enough for government work, actually. And it's not just flesh. Lots of things behave approximately like water, which is handy for all sorts of back-of-envelope estimates.
From the abstract:
"The Index captures technical exposure, where AI can perform occupational tasks, not displacement outcomes or adoption timelines." (emphasis mine)
The 11.7% figures is the modeled reduction in "wage value", which appears to be marketplace value of (human) work.
I assume they use N⋅D rather than ND to make it explicit these are 2 different variables. That's not necessary for 6N because variable names don't start with a number by convention.
Well no actually it'd still clear to me that they mean the the multiplication of 3 different variables X, N, and D.
I don't think of it as eliding obvious operators. Rather in mathematics juxtaposition is used as an operator to represent multiplication. You would never elide an addition operator.
So X next to D still means multiplication as long as you can tell that X and D are separate entities.
I would wonder why they switched conventions in the middle of an expression though.
Only cladistically. There's a better argument that there's no such thing as a tree or crab. As far as I understand, at least the common ancestor of all fish was still a fish.
"Only cladistically" is a bit like saying "only in reality" imo. :P
> As far as I understand, at least the common ancestor of all fish was still a fish.
Well.. eel-like probably more I'd guess. But yea.
The issue with "fish" is that people want it to be cladistic (trout and shark are fish) AND function (whales are not fish), and potentially also another anti-function (eels are not fish). You can't have it both/all three ways.
With crabs and trees that's 100% function, and that's fine imo.
> "Only cladistically" is a bit like saying "only in reality" imo.
In the sense that "imo" means "in my opinion, not necessarily in reality".
Clades are just a view on biology. It's not the only one. Otherwise, very few things in biology would exist.
> Well.. eel-like probably more I'd guess.
Still a fish.
The only people who want to see fish cladistically are the ones who don't want fish to be a thing. Fish are obviously a thing, and they are obviously not a clade. It describes function: water, gills, spine.
Mosh is excellent. It lets remote sessions survive (well, automatically and transparently recover from) disruption that reliably kills ssh.
I basically don't use ssh at all any more for interactive sessions, because I'm sick of a few lost packets on wifi or a weak cell signal dropping my connections and forcing me to start over.
Tmux, I used to use and eventually abandoned. I decided I didn't need two keyboard-based window managers (I use Spectacle on Mac) and the one that was only for shells was the one that could go. I have replaced it with nothing, so far, aside from that I just open more Terminal.app windows now (I also used to use iTerm2, for years, until it dawned on me that I was using exactly nothing in it that's not also provided by Terminal.app, and the latter's got better input latency, so I was suffering an extra installed program and slightly less responsive typing for no reason at all)
On mobile so I’m not sure which case OSC52 applies to, but I use mosh+tmux 8-10 hours a day. Both bracketed paste and tmux selection setting local clipboard work fine
Yes. Mosh is a seamless replacement for ssh, and screen is a mostly seamless replacement for tmux. One more level is Mosh+byobu, which is so useful I don't even bother with plain terminals most the time.
Haha I think you have the history backwards here. Tmux was created as a replacement for screen when it was 20 years old! Speaking as someone with ‘set -g prefix c-a’ in their .tmux.conf because my muscle memory is so used to the screen hotkeys.
Hey maybe I do! Great to learn the history here. Regardless, my point is that Byobu is very nice compared to either. That may now be lost due to my careless comment.
https://en.wikipedia.org/wiki/Division_by_zero#Alternative_n...
Less sarcastically, ∞ as an extension to the number line which has the properties n / 0 = ∞ and ∞ > n for all n != ∞ is both in common parlance and can be rigorously formalized.
reply