To be fair, "tagging images", i.e. supervised learning, is pretty widely supplemented by unsupervised techniques today because it usually works a lot better. He also says things like just scaling up won't work forever, which seems to be true; the current LLM models are not just scaled up versions of previous ones, and the current LLM craze itself is the result of architectural innovation.

He also says things like "I argued that 'deep learning is hitting a wall'" which suggests he doesn't really understand the difference between deep learning, which involves the very fundamental observation that deeper computational graphs can be much more efficient than shallower ones, and specific examples like a particular LLM architecture.

Paragliders: the only aircraft that you assemble while you're taking off.

This guy was an adrenaline junkie. The very kind of guy who is attracted to aviation and the last kind of guy who should be practicing it.

Lots of assumptions. He completed a lot of flights, many of them more carefully planned than anything most of us have ever done, flying or otherwise. He died on what sounds like a pleasure flight and "fell ill while flying" might be Italian police speculation for "had a heart attack."

You don't jump off a balloon at 100,000 feet without a lot of calculation. Baumgartner was the spam in the can but there are whole teams behind these things, from figuring out whether it's possible to designing the spacesuit that kept him alive.

People have been getting caught sneaking around by low tech means for thousands of years too. Feeling up your subordinate in public has always come with some risk of getting found out, especially if you're rich and a somewhat public figure.

True, you can always have leaks, but hydrogen leaks seem to be way, way more common than leaks of other fuels as a percentage of launch attempts. The shuttle was scrubbed on average almost once per launch, and a large percentage of the scrubs were caused by hydrogen leaks (source).

And that's on top of the whole embrittlement problem, which can lead to catastrophic hardware failures if you're reusing parts over a long period of time, which is another reason why folks trying to do reusable rockets (e.g. SpaceX) tend to avoid it. And if you think embrittlement is a risk in something that gets used once, imagine the effect on fuel lines in cars that are pressurized for decades at a time.

It is a really, really nasty fuel, IMO. Mind you, hydrazine is worse in some ways, but that doesn't mean hydrogen isn't nasty. :-)

YES! but most data comes back... thing is, long term backups don't exist or the immediately important data is recent not old. So you may corrupt 6 months of backups and they have last years offline backups but those end up not being that valuable.

Say you have a cloud backup service... if they gain account access, they can go into that account and trash it all.

One possible exception: Rental cars at airports. But only to a point.

You're grossly overestimating the required range for most of those things. School buses travel an average of just 63 miles per day. Mail trucks? 25 miles per day. Or were you planning to charge them only on weekends? :-)

Mind you, that's not the whole story for school buses, because they also have to be able to take the sports teams and bands to out-of-town games and competitions within a 200-mile radius or thereabouts, so for those long runs, you would need at least a 400-mile range to do a round trip, or else the driver would need to be able to take it to a Megacharger or similar close to the destination with a 200-mile range. But you could also keep a small number of diesel or gasoline buses for that purpose if you want, at least for the foreseeable future.

What I don't get is why people will buy stable coins that do not get paid interest like a bank does. I've had some for a while as part of a larger transaction but I was given interest at a good rate for the time.

To be clear, I'm not saying it isn't the right choice for them. I'm just bothered by claims that they don't have a choice, as though this were forced, rather than a strategic decision to prioritize profits and simplicity of their billing architecture over inclusiveness.

Sure. That's common when technology is in its infancy. Those problems could be solved, given enough time and money. In theory, so could the leaks, though given that NASA tried to fix them for three decades and still had leak problems on Artemis, that remains a theory. :-D

Hydrogen has much bigger problems than filling time, freezes, or the occasional explosion from dispensing a gas at high pressure incorrectly. If those were the biggest problems, I'd be totally in favor of doing more with hydrogen and fuel cells, because those are manageable and/or fixable technical problems.

A much bigger problem is fact that almost all hydrogen comes from natural gas, which makes it anything but green. The dirty little secret is that your losses from leaks would make distributing it over oil pipeline infrastructure completely infeasible, so you'll end up distributing natural gas instead, and cracking it to make hydrogen. And now, you have all the CO2 emissions from burning natural gas, plus all the efficiency loss from cracking the natural gas ahead of time, and you're about as green as a forest fire.

And of course, if you get it through electrolysis instead, you're likely wasting considerably more than half the energy you put in, versus more like 1% loss when charging a battery, making it a huge drain on our power grid.

And the elephant in the room is the cost per mile. In California, current prices for hydrogen fueling are hovering around $36 per kg, or about 50 cents per mile in a typical hydrogen-powered car. This makes it almost an order of magnitude more expensive than BEVs, and there's no real reason to believe that this will improve at this point, given how long it has failed to improve. At best, the whole thing feels more like a glorified government subsidy capture scheme, rather than a serious means of powering cars.

Hydrogen is a terrible idea on multiple levels. The fact that ostensibly it could have been a short-term workaround to provide multiple means of getting energy into rural areas (gas pipelines and power grid) is nice in theory, but in practice, the losses are just way too high, and batteries just work way too well, so the benefits just don't hold up in practice.

*shrugs*

Hydrogen leaks were one of the most hated aspects of the Shuttle design, too. It's routine until it doesn't work. Then it's a huge pain in the backside. The shuttle, Artemis... both of the two big NASA designs that used hydrogen had big headaches from hydrogen leaks. At some point, you start to see a pattern. (Whether that pattern is hydrogen or defense contractors, I couldn't say. :-D )

Sorry, yes. Forgot that 24 / 360. Skip to "within a few seconds is fine."

It's fairly easy to take a sight within less than a second of the intended time with a bit of practice and an assistant to call the mark. When you actually take fixes you take several sights, ideally at large angles to each other, and random errors average out. For manual navigation you can get away with two, but you generally take three, which gives you a triangular intersect, and you plot your fix at the centre of the triangle. That's typically fine for ~1 nm under good conditions, with practice.

An automatic system can measure the angle more precisely, potentially take many sights simultaneously, and do it a the time resolution of whatever clock you're using. If you get to the point where you actually need to be worried about your crystal clock's accuracy, you can get one of these:
https://www.microchip.com/en-u...

*shrugs*

There are literally almost one thousand companies that do this. There's no way that none of them would be willing to process your payments for a high enough fee.

There are quite literally payment processors that specialize in high-risk merchant accounts (e.g. PayCompass).

Yeah, there have been problems over the years caused by payment processors being a**holes. And other companies have come in to fill the holes they left behind. That's the thing about capitalism: When one company won't do business with you, there's almost always somebody else who will, if there's enough money in it. And there's *definitely* enough money in it.

These companies just don't want to fool with it. And I get that. But I really get tired of them acting like they have no choice. They have a choice. They made the choice to pay lower fees on everything in exchange for not selling that. They could also have chosen to use a processor that specializes in high-risk transactions and raised commissions across the board to make up for it. They could also have split their purchases into adult and non-adult purchases and used a high-risk merchant account for the first and a normal merchant account for the second, and raised commissions only on adult content that the first processor considered high-risk, adjusting the line over time, as needed.

There are always options. Those who say they have no choice are almost always just making excuses for poorly thought out decisions.

The main problem is probably that Steam supports PayPal, which is one of the most restrictive payment processors out there, and PayPal probably gets mad if you can't pay for certain things with PayPal because of PayPal's AUP. Heck, Steam might even use PayPal under the hood for all of their transactions, for all I know. But the thing is, Steam chose to tie themselves to that payment processor, and nothing stops them from walking away.

And by "more difficult", you mean that there is a massively larger amount of efficiency loss. If you're lucky, you recapture 60% of the energy in the fuel, and multiply that times the 75% efficiency for electrolysis, if you're lucky, and you're throwing away more than half the energy that goes in. So if folks are freaking out about the power grid not being able to handle EVs, imagine what would happen if you doubled the power requirements.