They expect to hit $300 by 2030, not $500. And comparing it to the cost of coal without CO2 capture makes no sense; you need to compare to the cost of alternatives to coal that don't have CO2 emissions but can still serve the same baseload role.

But yes, it does seem like too expensive of a solution. They'll need to do much better than $300.

By the numbers in TFA, we would need 2000 of such plants to hit 2030 climate targets

The numbers for doing anything sounds large when you apply it to the whole planet. But the planet has 8 billion people. Exactly how many man-hours of their labour, for how long do you think was needed building this plant and producing the raw materials that it was built from? The answer is going to be a tiny number relative to the total man-hours of the global labour pool.

Everything we do on a global scale involves mind-boggling numbers. It's so dumb how people act incredulous at large numbers when we already deal with large numbers with power generation, oil production, steel, fertilizer, on and on and on. The scale of the infrastructure we've had to build for thee things is staggering. "2000 plants"? I don't know how to break it to you, but there's an estimated 10 million factories on Earth.

Playing "these numbers sound big!" isn't a useful way to analyze the situation. You have to look at cost effectiveness vs. alternatives. If it's cost effective, said 2000 plants (or more realistically, fewer, larger ones) will get built. If other options are more cost effective, they'll happen instead.

The math is irrefutable

Your understanding of what's happening is not.

Deep buried rocks are not in some minimum energy state with respect to the atmosphere. As you can readily see from the rapid chemical changes that happen to lava flows when exposed to the atmosphere - most visibly, the oxidation of FeO to Fe2O3. Not only is deep buried rock not exposed to the atmosphere, but the thermodynamic equlibria are also different under their different temperatures and pressures. In said rocks, calcium and magnesium oxide and hydroxide - as with a typical basaltic lava flow - like 15%-ish of their mass in basalt like that under Hellisheiði - can bind with CO2 to form carbonates. This happens naturally (calcite, aragonite, magnesite, dolomite, etc), and we can induce it artificially as well by providing the CO2. And it most demonstrably works - on surprisingly short timescales, too. But you need the right strata to be present, and you need wells down to them.

If you don't believe me, quick question: you know what quicklime is? You know how you make it? You take limestone (calcium carbonate), and you apply heat to it to decompose it to calcium oxide and carbon dioxide. You have to input energy, even at surface conditions, let alone under the pressure at depth, to turn it into calcium oxide. It's more thermodynamically favourable for it to be in the state of calcium carbonate than it is for it to be in the state of calcium oxide plus CO2.

You can potentially see this in your own house. Ever seen a concrete slab that's started to burst out, exposing the rebar? Do you know why that happens? Carbonation. Steel in the high-pH environment of concrete is rendered passive; oxidation rates are almost nonexistent. However, concrete contains calcium hydroxide. This slowly reacts with carbon dioxide from the atmosphere to form, again, calcium carbonate. Because the very cement in the concrete was itself made by applying energy to drive the CO2 off, and it's now returning to its earlier state. As the carbonated layer gets deeper and deeper, it eventually reaches the steel; the pH drops, the steel is no longer protected, starts to rust, expands, and the concrete spalls out.

Thermodynamic equilibria vary with temperature and pressure - that is, a reaction may go in one direction in certain conditions, but the opposite direction in others. For example, driving CO2 off calcium carbonate with heat to make quicklime, cooling it down, then react with water to make calcium hydroxide, which slowly reabsorbs CO2; at elevated temperature at 1 ATM, the opposite reaction was favoured to at lower temperatures at 1 ATM. The balance is complex, because heat may favour a given reaction but pressure may disfavour it, or vice versa; it all depends on the specifics.

Deep rocks are simply not in thermodynamic equilibrium with the surface. Period. Because this is a system with huge differences in temperature and pressure with depth, changing their thermodynamic equilibria as matter cycles both up and down through the crust and mantle. Gas commonly escapes during these changes, rising to the surface and is lost. Rocks that would slowly absorb CO2 (or oxygen) if they were on the surface to exist in a non-reacted - but reactive - state, because the things they could react with simply don't exist there. A situation you can remedy by supplying said CO2.

Do you understand now?

it would reduce atmospheric carbon more if this green energy was used to power whatever we want to power

This green energy is located in Iceland, a remote island country where essentially 100% of all electricity consumption is already from clean energy. The only way we can export energy (barring the megaengineering project of the world's longest subsea power cable, by huge margins) is to do energy-intensive industrial activity here. For example, there's not a single bauxite mine in the country, but we're a major alumium producer, importing bauxite and exporting the metal, and in effect, exporting its embodied power. It would be more efficient to do the processing closer to ore production, but the power is here, not there, and we can't just send it there.

It's not "a random place". It's at Hellisheiði, Iceland's largest geothermal plant, and one of the largest geothermal plants in the world. So that they can inject the CO2 down the wells of the plant, where it binds with the rock to form carbonates.

1) Alle Dinge sind Gift, und nichts ist ohne Gift; allein die Dosis macht, dass ein Ding kein Gift ist.

2) Increasing CO2 levels does help plants, via reducing photorespiration per unit carbon fixed. Not, however, as much as killing them with worsened weather harms them (in particular; a warming client sees the monsoon belts move poleward, dries out soil faster, makes rivers more seasonal, and increases the intensity of peak rain events - aka, both drought and flood become more common). Plants also have optimal cultivation temperatures, and most are C3 plants, which tend to not like hot weather. Higher temperatures make them less efficient, and again, to a greater degree than CO2 helps them. C4 plants are generally better at dealing with drought and higher temperatures, but they don't benefit as much from increased CO2 availability, as they're already so good at capturing CO2 and could grow in CO2 levels a tiny fraction of that which we have now.

3) This is a bizarre argument. So, say, if I dump tonnes of cobalt in your drinking water, that's not pollution, because the human body needs to consume billionths of a gram per day? Some bacteria produce energy from oxidizing arsenic or using arsenic compounds to conduct photosynthesis - you okay with me contaminating your food supply with it? Some bacteria consume uranium - okay for me to fill your air with uranium dust?

The post you're responding to is literally quoting the dictionary.

What I care about is, does it come with CUDA preinstalled and everything ready to use the vast number of PyTorch-based packages?

You say in response to a person who uses it constantly.

I used to be peripherally involved with a legacy system that had, with some effort, been converted recently from 8" floppies to those new-fangled 3.5" gadgets. It used a unique file system that required a particular format. I figured out a way to use Linux to format the floppies and write disk images. We noted a sharp drop in floppy quality after about 2008.

The system used more modern storage at runtime, but making it boot off something more modern (e.g. USB) would have required a boot ROM upgrade. This was within our technical capability - the people who had written the original boot ROM were long gone by then, of course - but it was more cost-effective to scrap the system and make something new to replace it. So we did.

...laura

Red Hat has launched Red Hat Enterprise Linux AI (RHEL AI), described as a foundation model platform that allows users to more seamlessly develop and deploy generative AI models.

Hopefully whatever they do there will rub off on Fedora. It's always a massive pain to get setup...not least of which because NVidia usually develops one or two GCC versions behind (and you can't use incompatible versions with nvcc), and by the time they're caught up, the distro is already end-of-life.

Probably just as accurate.

Hopefully people will learn:

"The cloud" doesn't mean mean "someone else's computer" it means "everyone else's computer".

Oh, they absolutely are available. Tons of uncensored models on Huggingface. But you have to run your own model or find a place that hosts said models.

Unfortunately it could also tell you where to find prostitutes and how to kill yourself...neither of which can be monetized today.

Well, TODAY maybe, but...

To all these "AI simultaneously sucks and will also destroy the planet through overuse" - exactly who do you think will be buying all this power and all these GPUs to use said power, if there's no economic value to it and a competitor could provide as good or better of a service without said insane expense?

The electricity generation market was valued at $1,6 trillion per year in 2023 (just generation, not distribution, grid services, etc). If you're going to be meaningfully increasing that, you're going to need to have some sizable percentage of that in added economic activity to justify it. And then atop that, the even more expensive aspect of said datacentres to consume said power. How many hundreds of billions or trillions of dollars are you positing that the market will decide is worth spending, while simultaneously having products that non-AI competitors could perform as well or better than? Like, adding a whole new "entire GDP of Russia"'s worth of electricity consumption and datacentres for something that consumers are indifferent to, is that what you think the market is going to pay for?

The reality is: you can run Phi-3 on a bloody smartphone without any sort of AI accelerator aboard it, and so long as you steer clear of trivia questions (or redirect them to RAG), it's excellent.. LLaMA 3 IMHO outperforms ChatGPT in most tasks and can run on a good gaming GPU. ~5 second generation = the power consumption of playing a video game for ~5 seconds. These are not world-eating levels of power consumption.

And the efficiency level for a given set of capabilities is growing exponentially - both exponentially on the hardware side and on the software side simultaneously, with a very fast doubling time. Now, we can of course offset this by using exponentially better models. And sure, in many places we will. But for any given task, you hit a point where its capabilities are good enough for the specific task it's given. Wherein, you're going to choose to apply those exponential gains to "exponentially cheaper and more efficient" rather than "exponentially more capable".

The other side of the coin is training new foundations (not finetunes, that's easy). But again, you and your investors have to believe that there's an economic case on the other end that will pay for the investment. Want a trillion dollars per year worth of training resources? Better have a clear path to tens of trillion dollars in revenue coming out the other side. That's just not happening. And again, training, too, becomes exponentially easier over time for a given level of capabilities.

Drop by user-created models at say civitai at some point and start browsing. We're talking user-created stuff, not things made by megacorps.