British Thermal Unit doesn't give any hint at all...

I suspect starting with power (vs. energy) is historical. It's horsepower, not oats/second.

But really, which unit is most natural depends on what you're engineering. If you're making a transmission line or a motor or a heater, power is what matters. With a car or a train, power is what gets you over the hill. And you don't buy fuel by the MJ, you buy it by the liter (or, gallon in Freedom Units).

Even just with electricity, storage as electricity is a relatively new thing outside of niche cases like battery backups for telephone substations (you know, back when it was 42V over a copper pair to your house). If you needed to store power, you did it as fuel for a generator. And then that has to take into account the efficiency of the generator.

Amazon has 1.57 million employees - about the same as the population of Philadelphia, PA. Philly had over 150,000 crimes last year, including 222 murders. It would be surprising if Amazon didn't have some level of crime (more than just jacking up the price of Subscribe-n-Save items).

That doesn't tell us anything about how frequent that crime is, or how frequently they get caught, or why those workers commit crimes. We don't know if those workers are doing it because they're underpaid, disgruntled, have a gambling addiction, or are paying expensive hookers.

We know that people who have lots of money and power are still willing to commit crimes to get more. (And we know they're paying expensive hookers...)

So this doesn't give us any real evidence Amazon is better or worse than companies we want to like. For that, we'd need to see better numbers. Which neither Amazon nor our favored companies want to share, for obvious reasons.

1 Joule (J) of energy = 1 watt of power for one second. 1J = 1Ws.

Then if you want time in your power, write 100 J/s instead of 100 W.

Interestingly, strobes (camera flashes) tend to be rated in Ws. A small strobe might be 50Ws, my big handheld strobes are 150Ws, and studio strobes can be 400Ws. Or, 50J, 150J, and 400J. The peak power during discharge is pretty incredible, though. They'll dump all that in 1/500s, so a 150Ws strobe has a peak power output of 75kW (!), similar to an EV motor.

Most lithium cells can discharge at 100 C (that is, at a current which would drain the cell in 1/100 of an hour)... for less than a minute.

This is how those book-sized car jumper batteries work. It's literally 4 banks of lithium batteries connected together and to the clamps via a power FET or relay and a bit of monitoring circuitry.

I'm sitting here looking at a single 5000mAh 21700 cell from one of my flashlights. 5Ah * 3.6V = 18 Wh. But it can discharge 1800W (!) for 1/100 of an hour.

So, is "17,703.5 MW of battery capacity" 17,703 megawatt-hours of capacity? Or is it 177 megawatt-hours of capacity, but it can drain in 1/100 of an hour for grid stabilization? Or somewhere in between? Who knows?

If you're buying a battery backup system, and you need 1800W, you would be very disappointed to discover it had a single 21700 cell in it... and would keep your stuff on for 36 seconds. Of course, if you bought it on Alibaba, you should have expected that...

MW is a unit of power. MWh is a unit of capacity - that is, power * time.

If it helps, think of it this way:
- Power is how frequently I can give a f*ck.
- Capacity is how many f*cks I have to give.

Linux release Caffeinated Cougar has entered the chat...

SystemD just wants to know when you're old enough to hit on. The next field to be added will be relationshipStatus.

I was mostly amused. I worked in firmware and secure boot for decades, so it wasn't really going to get away with it on my website.

But it has convinced me I will never upload any personal information to anything that looks vibe-coded. Because those people have no idea what their AI-generated code is doing. And there's a wide spectrum of bugs in between "really working" and "so preposterously f'd up that a newbie would notice.

It matters, for two big reasons:

1) It tells you whether you're benchmarking something that's actually useful. If you're benchmarking basketball by how many times a player bounces the ball on the court, that might indirectly correlate to how well they score (since it's an indicator of time with the ball), but it leads you to optimize the wrong thing.

2) It tells you when the model is good enough to be useful for real-world tasks. "Our previous smoke detector was only rated 1.3 flugelhoffers; the new one is 1.6 flugelhoffers!" Yes, but what are the odds it actually detects a fire of the type I might have in my house? Does that require a sensitivity of 1 flugelhoffer, and both are good enough? Or does it need 100 flugelhoffers, and we're nowhere close?

So, funny story. I just implemented Cloudflare Turnstile on my website, because I'm tired of all the AI bots scraping it.

I decided to see what Github Copilot would propose for an implementation. Its first pass wasn't a good integration with my site, but it did show me what pieces of the puzzle I'd need. Took me a few hours after that to reimplement it cleanly. I figure that first pass probably saved me half a day of futzing around myself, though.

But then it wasn't working on third-party redirects (like FB links); it'd just refresh the widget forever. Took me a couple hours to find the one line fix for the problem. And then I thought, hey, let's see if Copilot could have found the fix. And its fix was both hilarious and terrifying:

1) If the page is refreshing, just hide the widget from the user.

2) And then store a no-permissions plaintext cookie on the client browser which says, "This site already passed turnstile, so accept it".

3) And while we're at it, disable third-party protection on the PHP session cookie.

I mean, yes, that does technically solve the problem of "the widget keeps refreshing and doesn't let users into my site". But only by defeating its security model and opening another security hole. It's like, "Copilot, I keep forgetting my safe combination." "Oh, I can help with that. First, wedge the door of the safe open so it can't close. And just to be sure, put a post-it with the combination on the side of the safe."

...tracking international shipments of sensitive technologies.

That's the point.

It doesn't matter if the tracking is effective in stopping the chips from reaching China. And the chips don't even need to be in China for them to be run by the Chinese; strangely enough, datacenters can be connected to the internet.

But every chip will need to pay the tracking tax.

For example, a new grad with a BS in Biology? Or a mid-career researcher?

And with what time limits? Is the amount of work in this benchmark something that would take the human a day? A week? A month?

I'd also like to know how quickly a new grad or mid-career researcher can identify which things the AI got right? For example, day it's asked a week's worth of work and gets 36% right = 14 hours. If it takes the human 10 hours to figure that out, it's a win. If it takes the human 20 hours to figure it out, it's not.

And how well could the human figure out ahead of time which things it thought the AI would get right? If the human only asks that subset, then the payoff is better. Say the human only asks the AI to do 20% of the tasks (8 hours of work), but now it takes 20% of the time to grade (so instead of 20 hours, it takes 4 hours). Now it's a win again.

Without knowing these things, it's like saying, "AI sucks at playing golf!" Without saying whether it's having trouble with 400-yard drives or just getting the ball into the windmill before the ramp goes up.

So, workers don't know what management is doing. Management doesn't know what workers are thinking. And a company named Blind of all things is going to help with that? Please tell me they have internal (or heck, external) marketing presentations titled "Blind Is Leading The Blind Leading The Blind"...

(Or is their internal leadership oblivious to this opportunity for marketing excellence? In that case, maybe the blind are leading Blind leading the blind leading the blind... It's turtles all the way down...)

Why are the thrusters running during the 18 months at Mars? Mars has a much thinner atmosphere, so at any reasonable altitude any spacecraft in orbit will encounter much less drag than on the ISS around Earth. The ISS does not run its thrusters continuously; it gets an occasional boost from a Soyuz. And there isn't much space debris around Mars compared with LEO, so not much to dodge. If the entire spacecraft lands, it's definitely not running its thrusters the entire time.

So that's 12-18 months (1 - 1.5 years) of operation, only when it's headed to Mars or back. Half the quoted amount.

Do they mean it has some sort of optics in front of it so that my eye focuses on the display as if it's at 20'+ distance, as it would subjects in a traditional mirror? If so, yes please. And please put that on the dashboard display, too. My 50+ year old eyes don't focus inside of 36" away, and my bifocals stop at 24". Aaaaand, the dash is between those two. Fortunately, I also drive an older car with analog gauges, and it's pretty easy to see where the dial points.

Otherwise, I have no idea what they mean.