yeah the reported range was 3x-94x.

obviously the 94x was on some weird edge case that happens to hit the sweet spot of their optimization. it might even be some 16x16 video they created as a test input for their pipelining.

most people would just ignore it and focus on the 3x which is still good, but you do you.

It's just getting sad in the cryptobro world.

Anything pegged to the US Dollar is effectively ... a US Dollar. The dollar depreciates every day and so does anything pegged to it. That's why people invest in other instruments (like stocks, bonds, money market accounts, real-estate, etc.) to keep the value up as the dollar value decreases.

It doesn't matter if you call it inflation (the dollar is worth less) or "shrinkflation" (you can buy that same pack of hot dogs but now is 8oz not 12oz) or whatever. If you keep your "investment" in USD under your mattress or wherever it won't reduce in its amount but it WILL reduce in what you can buy with it. That $15K car is now $30K. That $30K SUV is now $70K. That $90K house is now $300K, etc.

So Yet Another StableCoin is Yet Another Way to yoink funds from real dollars (not a great "investment") to fake things that are magically linked to the dollar, until the SEC says it isn't or the cryptobros tell you they went belly up and you own a whole lot of math worth nothing.

But hey, it's a news story. About soon to be losers.

Are you a victim? Would you like to be one? At least go to Vegas or Monaco or Macao where your gamble has a chance of paying off. This cryptobro shite makes everyone poorer.

It's not that simple.

Russia turned commie in 1917. By 1991, there was no longer any living memory of how markets worked. People had been born into corruption and cynicism. They were fed lies with their breastmilk. They knew nothing else.

China turned commie in 1949 and started opening up in 1978. Living people had been shopkeepers and factory managers who understood markets and supply & demand.

The level of corruption in China was much less and also different. In Russia, you pay a bribe to some bureaucrat to get your business license, and he steps out of the way so you can deal with the next official with his hand out. In China, they have the concept of guanxi ("connections"). You pay a bribe, and the recipient becomes your champion. He'll guide you through the process, easing things every step of the way. He'll even invite you to his home to meet his family. And, of course, he'll be available (for a fee) to help you with any other problems that arise.

I went to the demo and tried it.

It's nice, but it needs to be half the price and half the weight before I'll buy it.

I was frustrated because the demo was scripted and entirely focused on consuming content. I wasn't allowed to pop up an editor or try to do anything productive with it.

Nonetheless, I am a potential future customer.

Musk lied on his immigration application form, and his political adversaries are saying his citizenship should be revoked.

I doubt anything will come of it, regardless of who wins tomorrow.

The thing that really slows things down in modern assembly isn't "number of instructions" (as you mentioned).

The things you really want to look at are "number of memory accesses" or "number of missed branch predictions." Aligning cache accesses correctly can also be a big win.

Yeap, insertion sort is faster than quicksort for N < ~12.

Apparently you don't, otherwise you wouldn't say Cisco is secure.

The government could pay for the defense to hire a counter-expert.

I don't really see in the article where the science around the Menendez brothers has changed. I might have missed something, but I don't see it.

Compare it to the arson cases, where "expert" witnesses were lying and saying it was arson when in reality it wasn't.

There are several hundred farms in Singapore, nearly all in the northwest corner. They grow mostly fresh, perishable produce.

There are many marine fish farms in Singaporean territorial waters.

Architecturally, it's not a problem. You just encapsulate the assembly in a function, and use polymorphism, based on whether the CPU feature is available. For the platforms that don't have the ability, they will just run slower. Apparently 94x slower or something.

Bubble sort is faster than Quicksort for less than 8 items. That sounds like nothing, but then you realize many if not most sorts done probably have fewer than 8 items.

They're too busy, the compiler writers are working on breaking compatibility with doubtful optimizations.

More secure than Cisco? Are you trying to say, with a straight face, that Cisco is secure? Cisco doesn't believe in passwords.

So from your recommendation, I would say avoid Duo Security for anything that matters. Oh, whoops.