The experiment to observe this behavior is pretty simple though (Young's double slit), and it was conducted more than 200 years ago. The explanation came much later but it's not like the phenomenon was hiding somewhere.
It’s both ridiculous and quite amazing really. The hint that there is something less random underneath it that we just haven’t figured out (and lack the resources to explore at this time) is tantalising.
Even if there isn’t, the way it seems all based on the uneven flow of state over spacetime is deeply fascinating for someone who studies computing.
10,000 square meters sound suspiciously small for a datacenter, even more so if you have to account for supporting facility? Maybe a small one? it's just 100m by 100m, which is smaller than most Walmart Supercenter.
JSD is just symmetrized KL, it's the forward KL + reverse KL.
In reinforcement learning, usually what we want is to find the optimal action, i.e. action that maximizes the reward, this translates to the so-called "mode-seeking" optimization, which is the reverse KL.
JSD is slightly different to forward KL + reverse KL (which is unbounded, whereas JSD measured in bits is in the range [0, 1]).
One way to interpret JSD(P, Q): Associate the distributions P and Q with two target classes, respectively. Pick a target class based on a fair coin flip. Then sample either from distribution P or distribution Q, depending on the outcome of the coin flip. The JSD is the mutual information between the resulting mixture distribution and the target class.
Alternative intuition: Suppose we want to measure the correlation between a feature X and a binary target class Y. We have a tabular data set with two columns X and Y, whose rows correspond to individual samples. JSD is the mutual information between the feature X and the target class Y, but after we resample our data (rows) to ensure that we have a balanced representation of the target class Y. If we measure the JSD in bits, the quantity 2^(JSD-1) is the fraction of times X correctly predicts Y, assuming balanced classes.
I've done Stockholm-Oslo without stopping to charge in December in my model y long range. Didn't really do anything special either, just obeyed the speed limits pretty much. Most of the drive was on autopilot(not fsd) because highways are boring.
Had a pretty healthy margin too, I charged on the outskirts of town on the way home 2 days later.
Unsurprising, for a Ferrari. I suspect it's designed for performance and not efficiency. Atrocious mileage is par for the course in this segment (see the Veyron)
I've done Stockholm - Oslo on a single charge in early winter, which is almost exactly that distance, so I'd say it does! Even kept me nice and toasty along the way!
The actual number of the EPA range is imaginary, yes. But it's useful for comparisons.
But if we're talking about comparisons between two vehicles, the vehicle with a 122kWh battery and a 280 EPA range will go less far and is much less efficient than the vehicle with a 84kWh batter and a 300 EPA range.
If you actually use them you'll see that they are far from frontier models. They are much more cost-effective for what they are, but frontier they are not.
Thing is these models can also be a propaganda machine whether you run it locally or not. This is true no matter the origins. Chinese LLMs will never shit-talk CCP, and it will always give a rosy depiction of the Chinese government. It's perfectly understandable if companies don't want things like that. US/EU models have these problems too, but at least there are some ways to fight that: with a lawsuit or a megaphone on social networks. With Chinese models there is nothing you can do.
And all of them will learn how expensive and difficult it is to make good hardware that the majority actually wants to use. Meta learned this the hard way (Facebook phone, Oculus, Meta home devices, etc.), the same for Google. I don't think OAI and Anthropic have the capital and time to ride out the hardware loss. Google and Meta could afford their hardware blunders because it's not their revenue sources.
You're confusing those cartridges when fired from a rifle length barrel vs. a handgun length barrel. A 1.5 inch barrel on some pocket carried revolver is not going to send 22 LR at anywhere near the speed of a 16 inch barrel
I use Blind sometimes to check the TC of a company. Most of the posts/comments there are either stupid, sexist, racist, or all of them. But it does feel like most of them are real. Blind requires verification by company email for posting, which I guess eliminates most of the bots.
reply