ComplexSystems

7 ms·

151.
by ComplexSystems 1y ago
> They are much better at the thing they are good at, and there are some new capabilities that are big, but they are still fundamentally next-token predictors. I don't really get this. Are you saying autoregressive LLMs won't q
152.
by ComplexSystems 1y ago
I doubt they're happy about having to be up there that long.
153.
by ComplexSystems 2y ago
How are you using Whisper?
154.
by ComplexSystems 2y ago
This is also apparently true for playing Go.
155.
by ComplexSystems 2y ago
The way I learned this, at least, 6 dB is what corresponds to a doubling of amplitude, but human ears compress this so it's 10 dB that corresponds to a doubling of volume - this is how the "bel" was supposedly calibrated to b
156.
by ComplexSystems 2y ago
Yes, I probably should have said "listening." I edited.
157.
by ComplexSystems 2y ago
I don't think any such conspiracy or secret-keeping is required; one need not attribute any of this to malice which can be attributed to incompetence. There already exists a system, on everyone's phones, which is listening to audi
158.
by ComplexSystems 2y ago
I can't even begin to tell you how many times I've been randomly having a conversation with someone, only to be alerted to the sound of the Google Assistant suddenly responding to what we're saying. Something we said was inte
159.
by ComplexSystems 2y ago
None of this is in the realm of "conspiracy theory." We aren't talking about the Earth being flat. Your phone is very clearly listening at all times, and sometimes activates and sends that data to servers. Given that, we can
160.
by ComplexSystems 2y ago
What do you think "AGI" is supposed to be?
161.
by ComplexSystems 2y ago
"LLMs, without pattern matching, can only do up to about integer division, and while they can calculate parity, they can't use it in their calculations." - what do you mean by this? Counting the number of 1's in a bitstr
162.
by ComplexSystems 2y ago
"There is no evidence that LLMs are the roadmap to AGI." - There's plenty of evidence. What do you think the last few years have been all about? Hell, GPT-4 would already have qualified as AGI about a decade ago.
163.
by ComplexSystems 2y ago
That 30 bits was not literally intended to represent only the object's noun category, but even if it did, none of the additional pieces of information you would like to add are going to change this picture much, because what one would
164.
by ComplexSystems 2y ago
These letters are jointly distributed, and the entropy of the joint distribution of a second of "plausible" English text is much lower than the naive sum of the marginal entropies of each letter. In fact, with LLMs that report the
165.
by ComplexSystems 2y ago
On the other hand, the amount of actual entropic "information" that is processed when you identify a Rubik's cube as such may be nowhere near as much as you think it is, and most importantly, 10 bits may be nowhere near as li
166.
by ComplexSystems 2y ago
There is no problem with the proof except the assumption that the value in the limit is the same as the value at infinity. If we simply define pi(n) as a function from N U {inf}, which gives the value that "pi" takes at the nth st
167.
by ComplexSystems 2y ago
Why don't they just say it then? Who cares?
168.
by ComplexSystems 2y ago
For now.
169.
by ComplexSystems 2y ago
SAT turns up everywhere because it's almost universal kind of problem. Since it is NP complete, everything in NP can be transformed into an instance of SAT. Since P is a subset of NP, everything in P can be also be turned into an insta
170.
by ComplexSystems 2y ago
"Bayesian counterargument (in caricature form) would be that MLE frequentists just choose an arbitrary (flat) prior, and penalty hyperparameters (common in NN) are a de facto prior." This has been my view for a while now. Is this
171.
by ComplexSystems 2y ago
I am curious how this changes if we condition on it flipping in the air at least once. Can we think of this result as a mixture distribution of a fair 50/50 chance of it flips at least once, and a delta function that is 100% at the sid
172.
by ComplexSystems 2y ago
One answer is that this "initial phase" may have lasted a total of a few hundred years, which is a tiny blip on geological scales. Even with humans, it looks like it's going that way.
173.
by ComplexSystems 2y ago
The term for this is "transparency." A codec is "transparent" if people can't tell the difference between the original and the compressed version.
174.
by ComplexSystems 2y ago
That's from 2023, and it sure as hell isn't how people voted this time: https://www.statista.com/statistics/1535295/presidential-ele...
175.
by ComplexSystems 2y ago
Yeah, emotions like "I wish I could afford to buy food."
176.
by ComplexSystems 2y ago
I'll just point out that when you say "inflation improved massively," you are talking about the second derivative of price. You are saying that there was a positive change in inflation, meaning that the rate of change of the
177.
by ComplexSystems 2y ago
Your rewritten "response" has the same problems I am pointing out. To the average voter, it says 1. Biden is good and inflation wasn't his fault 2. Biden's handling of it was good, he did all good things, Biden is good 3
178.
by ComplexSystems 2y ago
No arguments there. I certainly expect tariffs will lead to inflation getting much worse.
179.
by ComplexSystems 2y ago
Well, for starters, a response that would have worked won't involve both of these contradictory positions at the same time: Position 1: Prices can never go down again unless inflation is negative and we get "deflation." Defla
180.
by ComplexSystems 2y ago
I don't think that the problem is that Democrats didn't explain the technical definition of inflation well enough. The problem is that people can't afford to buy things. Having better infographics on how inflation is the deri

More ›