It occurs to me that this is a good use of massive solar plants. It wouldn't cost much to idle your oxygen-separation equipment when the sun isn't shining, so you wouldn't need much in the way of battery storage. Grid scale solar without battery backup in a sunny area (like south Texas) can cost as little as $0.03/kWh, which would give you a separation cost of $4.5 to $24 per ton of LOX. Obviously, if you were producing LOX at a scale needed to fuel a fleet of Starships, you'd work to get that towards the bottom of the scale -- so the LOX loadout for a ship could cost on the order of 3500 * 4.5 = $15,750. To launch 150 tons to orbit. Of course you still need methane.

Could you make "green" methane (i.e. without using fossil fuels) with a big solar farm, and what would that cost? You'd do it with the Sabatier reaction to combine CO2 and H2 to get CH4. To make a ton of CH4 you need 2.75 tons of CO2 and 0.5 tons of H2 (stochiometry, dawg). To get a ton of CO2 with direct air capture takes about 2000 kWh of electricity, so 5500 kWh for the CO2. At $0.03/kWh that's $165 for the CO2. However, producing the half-ton of H2 with electrolysis would take 25,000 kWh, so $750. This puts the raw materials cost of green CH4 at around $915. The Sabatier reaction would add a little more, call it $930 in all.

So... Starship could be entirely solar-powered at a cost of around 3500 * 4.5 + 1000 * 930 = ~$946k, assuming $.03/kWh, ignoring equipment and storage overhead. It turns out that the cost is utterly dominated by the cost of methane production; LOX is all but free. But the cost of solar will likely continue to go down so... fuel costs could indeed get really, really low, even with a zero-carbon strategy. Perhaps as low as $2/kg to LEO.

Rapid Accelerated Disassembly?

Rapid Anomalous Disassembly?

Or did you mean R.U.D. (Rapid Unscheduled/Unplanned Disassembly)?

Starship burns methane, not RP1.

Between SuperHeavy and Starship, a fully-loaded stack needs 3500 tons of LOX and 1000 tons of CH4. So what do those cost?

Well, oxygen is easy to get from the atmosphere, so the cost of LOX is really just some equipment (which isn't terribly expensive to buy and maintain) plus electricity, and the cost ends up being dominated by the cost of electricity. It takes between 150 kWh and 800 kWh to separate and liquify a ton of oxygen, so if you're paying $0.10 per kWh, LOX costs $15-80 per ton. There are some other costs to handle and store it, so let's say $100/ton.

CH4 can be created many ways. The cheapest is probably to purify natural gas, which costs about $190 per ton (that site shows ~$5 per 1000 ft^3, and a ton is 38k ft^3). Add some costs for purification and cooling, so call it $250/ton.

3500 tons LOX * $100/ton + 1000 tons CH4 * 250/ton = $600k. Musk usually calls it $1M, which seems pretty reasonable, since they're probably not separating/purifiying it themselves and there transportation costs. 150 tons of payload to LEO with $1M worth of fuel means the fuel-only cost is $6.67/kg.

Nonsense. Our only experience with reusable orbital rockets is the space shuttle, which was an unsustainably-expensive and complex beast that was more refurbishable than reusable and had a payload one fifth of what Starship is designed for. It's all of the differences that aim to make Starship both reusable and cheap that make it hard. It's possible that it's just too ambitious, that we don't yet have the technology to make a cheap, fully-reusable (not refurbishable, reusable) orbital rocket with massive capacity. No one else has done it... no one else is even trying, that's how hard it is.

Failure is expected. If they managed to launch and land both Starship and SuperHeavy in less than a dozen test flights, that would be the surprise.

Nothing. SpaceX is doing fine. Starship is ambitious. It is also being developed in a manner not suitable to the sensibilities of the Western aerospace commentariat. SpaceX performs many tests, analyzes many failures and refines designs accordingly. This produces great designs at low cost, in less time, and many dramatic RUDs. The Russians did the same. They performed many tests on initially flawed designs and fixed the flaws they discovered until they had confidence in their designs.

The traditional Western, big aerospace way, as we can clearly see with SLS is to take a decade or more and consumes oceans of money analyzing a paper design beyond any conceivable failure mode. This works, but it's extremely expensive, glacially slow, and suitable only for national superpower scale budgets funding cost plus contractors with little to no thought given to a feasible long term business model. That's why all their marquee designs are now historic, and the next one is still nascent, wildly over budget, years late and likely redundant.

So don't worry too much about the deep thoughts of our professional spectators. You can be absolutely certain that Musk doesn't.

You'd think Broadcom et al. would pitch a fit given the billions they've spent developing Wi-Fi standards that include 6 GHz, developing 6 GHz devices, etc. It's not just owners of existing equipment. It's an entire industry that has been investing in 6 GHz Wi-Fi for years now.

I'm surprised, and you should be too, if your view is evidence-based, because this is a new effect that was not predicted by any of the previous models, which already consider the melting permafrost, methane emissions and fossil fuel use.

I am saying I have no patience for the drearily predictable "quality" and "safety" FUD. There are severe problems in healthcare. Bad enough to risk neglecting our worship of medical authority. Bad enough to risk suffering possible unknown failures as an alternative to our chronic known failures.

This presumes we have quality. Do you believe that, without doubt? I don't. I have a lifetime of anecdotal evidence of failures by doctors, personally and among family, friends and others. Without (hopefully) inviting a deluge of corroboration, I can assure you the people reading this now can bury us in such stories.

Beyond that, we are in desperate need of lower cost solutions for medicine. You're free to attribute the extreme costs we see however you wish, but finger pointing won't fix it: the powers and interests involved aren't listening. What is needed is a disruption, and this looks like a real possibility. I, at least, don't immediately dismiss it with AMA FUD.

Show evidence of code theft, where these models are built with proprietary code that hasn't been liberally licensed and freely offered. Otherwise you're engaging in FUD.

A.) Zero ads: wading through prevailing search engines is a total shit show. I'm paying for LLM service, and I don't have to suffer that crap.
B.) Most examples are written by learners that are, themselves, ignorant of the sort of subtleties I mentioned. LLMs do better than that: they evaluate code and point out stuff that would otherwise be overlooked.

Yesterday, I wanted an example of a PIO program to generate high resolution, arbitrary waveform (variable frequency) PWM output using DMA on an RP2350 MCU. Gemini 2.5 Pro generated a correct, working, basic example. I refined it further by changing and adding requirements to deal with end state, corner cases and the deficiencies in the generated code. The final result works perfectly. Guessing here: It took probably perhaps 25% of the time to accomplish this than it would have without "AI." And while PIO appears simple, there is actually a lot of subtlety in the hardware that a new PIO programmer, without an AI, would likely either not know provides useful capabilities or would overlook, yielding less than optimal results and/or actual flaws. By using Gemini, I believe the code is on par with what a PIO expert would have produced.

So yes, it is actually helpful. And no, I don't believe this makes me or others like me obsolete: non-technical people cannot achieve the same results in reasonable amounts of time because they don't even know what to ask for, much less how to evaluate the answers.

All Overstreet needs to is continue development as he wishes. There is no fundamental reason bcachefs must be included in Linus's mainline kernel. The kernel has loadable modules. This work can simply be a loadable module. There are tools to make this next to transparent to an end user, up to and including as a root file system. ZFS On Linux has existed this way for 15 years now. There are entire commercial empires built around it, and it has never, at any point, been in the Linux mainline code base.

Whatever benefit bcachefs previously enjoyed by being in the mainline kernel has been entirely squandered by the recurring drama. Best to separate the parties and forego future squabbles. Another wise and brave decision from Torvalds.

Diminished maybe, but not all that much.

I think we can reasonably assume that if there's a huge blackout, it won't last forever. A lot of smart people will work hard on getting things up and running again. A few years ago in the USA it lasted for a bit longer, what was it, a week or two? Recently in Spain it lasted a few days. But all those power stations and power grid operators don't just shrug and go home. So getting through those days is probably all it takes for any reasonably realistic scenario.

And you can build things up piecewise. I've got my solar now. The next thing will be a battery. Once I have that, I can think about an electric car.

Oh, and just to add one more thing . Some network protocols just fundamentally can't work under NAT. Ex: FTP. So party of the reason this seems to worn is because we changed the way we design protocols to be NAt compatible. That's dumb. The Internet should just work.

I work for a company that makes large industrial machines. Some of the machines are, themselves, networks. They have motors and sensors and encoders and PID controllers and more. But too many of those devices assume ipv4, as do the corporate networks they live behind. So we have to assign all the embedded devices IPv4 addresses that don't conflict with the corporate s network, then apply NAT. It is overcomplicated, and so I can't remotely monitor the devices. To solve this there are a gzillion 3rd-party companies each with their own tools and APIs for remote monitoring.

Each device ought to be able to assign itself a unique IPv6 address and we could talk straight to it. Instead we go through a myriad of 3rd-party NAT hacks to get there.

A firewall is fine. Multiple levels for firewall is fine. Multiple levels of firewall each one rewriting the IP address is a nightmare. Often time today teams assume NAT is a firewall feature, when in reality firewall don't need NAT to function. It's just a hack.