Veedrac

I'd assume the design as proposed isn't overbuilt, so we'd need to go from, say, 4 months cycle times to 8 hour cycle times, which is maybe 350x.

While that's not going to increase the cost by 350x directly, it is going to change the character of the pile from a bunch of dirt to a bunch of dirty pipes. This makes a lot of the simplifying assumptions no longer work; like you can no longer ignore the heat losses through the rods, or the lower thermal mass of the rods.

And to be clear, you can do this. There are faster-cycling thermal storage solutions out there. It's just not implied from the claim that these solutions would be so much better than batteries.

I meant ‘bare minimum’ in the sense of what's needed for the system to be profitable, not in the sense of ignoring costs.

The problem with using this approach for daily cycled loads is that it relies on passive heat transfer to distribute heat through substantial regions of dirt. This simply doesn't work for daily storage.

You can overbuild, but then your energy losses are going to be immense, because you never saturate or drain the bulk of the material, and are just losing energy to it.

You could build faster cycling systems instead, and active systems especially can cycle reasonably fast, but then your dominant costs no longer reduce down to a pile of dirt with a few rods stuck into it.

500x should be conceptualized as the bare minimum, not evidence of massive low-hanging fruit.

A battery that cycles daily makes revenue on its capacity about 350 times in a year. A seasonal energy store makes revenue on its capacity about once in a year.

A battery arbitrages between the most expensive and least expensive energy generators in the system. A seasonal energy store arbitrages between seasonal price averages.

A battery smoothing out solar production is operating on the difference between how much sun there is in the day, and how much sun there is at night. A seasonal energy store in the same role averages between summer and winter.

A factor 500 cheaper plus a significant quantity of solar energy production is about where you'd expect this kind of thermal storage to start making economic sense.

The article never frames this as off-grid storage for individual homeowners. They say

> A typical site is a factory, power plant, or town with a large earthen mound at the edge. The mound might be the size of a house for a smaller factory, and up to many football fields for a large power plant. Surrounding the earthen mound will be high-density, low-profile solar arrays.

I agree that trying to give every suburban house its own rock pile would not be very practical.

> The pile of rock required to heat an average home for a couple of weeks handily outweighs the house itself.

Maybe I'm failing to follow the intended argument here, but I do not see what is expensive about this. Houses are hollow; they do not weigh all that much. Dirt is cheap, especially when sourced locally.

FWIW another way to handle this is a local function with early returns. In this example you can also just use a single regex and most likely win performance too.

:= is so common in other languages, or even other Python statements, that it doesn't bother me at all, but Python is such an imperative-first language that its presence is rarely felt.

There is no honor in hiding behind euphemisms. Rationalists say ‘low confidence’ and ‘high confidence’ all the time, just not when they're making an actual bet and need to directly compare credences. And the 16.27% mockery is completely dishonest. They used less than a single significant figure.

I'm not really following. You said, "I don’t think we can scale up storage enough at any reasonable cost." We checked the math. The math said that even with pessimistic numbers, that assume no cost improvement and that everything deprecates as fast as the batteries, the amount of storage needed was readily affordable. You agree that there's a reasonable chance the price could go down, and agree that parts other than the battery chemistry do indeed last much longer.

Isn't that the whole debate? What are we arguing about at this point?

You don't have to go all the way to battery prices scaling to 5% of the price they are today to consider it likely the price continues to fall at all.

Replacing the whole system every 20 years still seems really wild to me, especially in worlds where price doesn't fall past the cost of maintenance, because the only wear part is the battery, and batteries are modular. We could, sure, I just showed the cost is a fraction of today's energy spend, it just seems really weird not to do something smarter instead.

I find this hard to believe. Let's math it. Let's put grid storage at $200k/MWh, or about 4x current battery prices. Let's keep it at that level for sake of argument. Let's very pessimistically say we need to replace 50% of storage every decade, and again for sake of argument let's say replacing the batteries costs the whole 4x (eg. no reuse of material or converters or grid connections). So the math there gives $1T/y for storage.

Idk, it seems like even pretty pessimistic arguments with frozen costs and early retirements suggest this is cheaper than the numbers you gave.

> That is assuming we can hit that scale in the next couple decades and that nothing happens to push up costs with increased scale.

This strikes me as odd to posit about a technology that has had a steady 23% cost reduction per doubling over the last 6 decimal orders of magnitude.

3% GDP over a single decade is certainly not a trivial amount, but it's worth noting the comparison here is spending more than that every year forever.

Similarly, 100 TWh sounds like a huge number, and it is, but it's like the equivalent capacity of one base Model 3 per 6 people globally. It's a lot in absolute terms, for sure, but it's by no means a crazy unachievable quantity of battery for a family of 6 to use.

This is basically just not true. Energy is fungible and the grid is demand driven. Every watt of solar displaces a watt of something else. In many places like the US we've basically stopped building nonrenewable energy generation as it's no longer economically competitive. China has to date still built a lot of coal powered generation even as most of the richer world shut theirs for greener sources, since China rightly and justly prioritized bringing a billion people out of severe poverty, but even in China the tides are turning, and we're seeing indications that 2025 could be their peak coal.

Things are getting better.

Even if, for sake of argument, one outright denies the evident exponential growth in solar, a purely linear extrapolation of 2024's rate from [1] puts solar equal to today's coal output by 2042. Solar is fundamentally a factory product, so this is a wildly pessimistic case, just enough interest in the product to keep the lines running. If you believe solar will grow for even a few more years, but still declare that it should level off, it's the mid 30s. If you're willing to just fit the established trend, even that's a vast underestimate. The difference between which of these to believe is just how brave you are.

[1] https://ourworldindata.org/grapher/electricity-production-by...

> At some point I was even hearing the claim that digitization (e.g. GenAI) was finally divorcing the tight connection between economic growth and resource extraction

Kind of? Mostly it's a result of renewables. The US basically doesn't build new fossil fuel power any more. Not every country is on exactly the same point on the curve but they're all approximately following the same curve. Energy use has also decoupled which I think is what you're referring to, but I'm not sure the cause neatly decouples.

The author plot the input/output on a graph, intuited (largely incorrectly, because that's not how sufficiently large state spaces look) that the output was vaguely pretty, and then... I mean that's it, they just said they have a plot of the space it operates on therefore it's silly to ascribe interesting features to the way it works.

And look, it's fine, they prefer words of a certain valence, particularly ones with the right negative connotations, I prefer other words with other valences. None of this means the concerns don't matter. Natural selection on human pathogens isn't anything particularly like human intelligence and it's still very effective at selecting outcomes that we don't want against our attempts to change that, as an incidental outcome of its optimization pressures. I think it's very important we don't build highly capable systems that select for outcomes we don't want and will do so against our attempts to change it.

No. You're not being civil right now. You are lacing your reply with insults.

Ah, you're right, thanks for the correction.

A defense of the Luddites against sewing machines was honestly not something I expected. Support for the Luddites in general, sure I can predict that, but the claim that sewing machines are bad, exploitative, produce "shitty" cloth, serve only the pursuit of profit...

I appreciate the good these organizations do, but I don't think that's the right measure of it. A person wouldn't in expectation serve global health better by becoming Catholic than by joining EA. That Catholicism is large isn't the same as them being effective at solving malaria. EA is tiny relative to the Church but still manages to support within an order of magnitude the funding you mentioned here, with exact numbers depending on how you count.

Similarly, it's not like government funding is an overlooked part of EA. Working on government and government aid programs is something EA talks about, high leverage areas like policy especially. If there's a more standard government role that an individual can take that has better outcomes than what EAs do, that would be an important update and I'd be interested in hearing it. But the criticism that EA is just not large enough is hard to action on, and more of a work in progress than a moral failing.

Please be civil.

I genuinely don't understand how you can point to someone's calibration curve where they've broadly done well, and cherry pick the failed predictions they made, and use this not just to claim that they're making bad predictions but that they're slimy about admitting error. What more could you possibly want from someone than a tally of their prediction record graded against the probability they explicitly assigned to it?

One man's modus ponens, as it goes.

I wish it were, but it's clearly not enough. There are plenty of people with healthy emotional empathy in the world, and yet children still die of easily preventable diseases.

I am plenty happy to simp for the Gates foundation, but I think it's important to acknowledge that becoming Bill Gates to support charity is not a strategy the average person can replicate. The question for me is how do I live my life to support the causes I care about, not who lives more impressive lives than me.

If putting up evidence about how people were wrong in their predictions, I suggest actually pointing at predictions that were wrong, rather than on recent predictions about the future that that you disagree over how they will resolve. If putting up evidence about how people make excuses for failing predictions, I suggest actually showing them do so, rather than projecting that they will do so and blaming them for your projection.

It seems very odd to criticize the group that most reliably and effectively funds global health and malaria prevention for this.

What is your alternative? What's your framework that makes you contribute to malaria prevention more or more effectively than EAs do? Or is the claim instead that people should shut down conversation within EA that strays from the EA mode?

Most of the benefits of manned space exploration are fake, yes. The incentives of the people doing it though aren't to be honest about it. NASA talks about how critical a constant human presence on the ISS is for ‘science’; never once have I heard them even try to justify $100B spend as more than a keyword.

There are a few real things a manned space program does: showing force in rocket technologies, serving as inspiration in STEM fields, and taking an early step to humans flourishing throughout the universe. There are also real but bad reasons, like NASA bloat is seen as a way to buy voters.

Trams don't actually scale higher than busses; the highest ridership BRTs have far more ridership than the highest ridership light rail. The key thing that makes it work is having a dedicated right of way. I expect busses get a bad rap as a scaled transit solution mostly because they have to share the roads so often. But it's indeed an advantage of light rail that it's a lot harder to make that mistake with it.

This is one of these sets of information that don't seem to make sense until you fit it all together.

Busses aren't much more efficient when riding down a lane than lower occupancy vehicles, but streets aren't bottlenecked by their roads, they're bottlenecked by their intersections. The key advantage of a bus is at the intersection. A bus holds the intersection for far less time than the equivalent passenger capacity of cars.

The problem bus lanes try to solve is dominantly that without them the traffic advantage of people riding busses mostly goes to people not riding busses, and this makes for a pretty terrible incentive structure. Busses are intrinsically disadvantaged against cars (schedules, uncertainty, routes), so if you don't help them, then people will prefer to drive. Bus lanes internalize the externality.

It's frustrating seeing Block 2 go through pain we thought we were past, but we're talking about flights this year January, March, and May. The last Block 1 flights were on the cusp of success, so we know if the tiles can get a bit more robust then the mission design works. We've seen Super Heavy do a successful booster catch. We've seen Starship do an orbital reentry all the way down to a soft landing.

Execution matters and this last half a year hasn't been it, but the evidence seems to me to show the theory can work, and that SpaceX isn't running out of capital or conviction.

An explosion can be fine, but the details really matter. The yardstick here is whether this iteration was better than the last, and the pace of progress if so.

I don't think this flight reassured anyone that the Starship program is making the improvements we'd expect.

Process isolation only works across processes. You can't just execute an untrusted block of code without setting up a whole sandbox for it.

Without speaking to all of the issues, this is all made much harder by the underlying hardware having extremely bad defaults. The idea that running code on the hardware is itself an unsafe operation means that any time you want to touch it you need proxies and intermediate languages and all this by default.

It's pretty easy for me to imagine a world where running code was safe by default, and this followed all the way to the top. It's obviously not that onerous, else JavaScript wouldn't be as successful as it is. Most of the details the post touches on are then just package management and grouping concerns.