In practice, storing energy for a longer period of time is basically never done, with the only real exception I can think of being space travel. And it's not how long the storage period is that matters. It's how quickly you need to get the energy when you're done. Sure, if you store a year worth of energy in a day and dribble it out over a year, a tiny fuel cell and a huge tank is great for cost. But literally space travel is the only practical application of that. For every real-world application other than space travel, you need to be able to dump the entire contents of the fuel cell in at most maybe five to ten times the period of time over which it was built up, if not less. That means either big fuel cells or a lot of fuel cells.

See above. And to that, I would add that converting electricity to hydrogen with electrolysis of water and back is likely to result in a loss of somewhere around 60% to 70% of the energy that you put in. So even if you somehow manage to find some rare edge case (e.g. trying to do solar in Alaska or something similarly nuts) where you really do want to store power long-term and spread it out over a long period of time, the loss of energy is still going to be around 5x as high from fuel cells as lithium ion batteries even factoring in the self-discharge rate over several months.

And that's before you factor in the additional losses from having to pressurize the hydrogen, which adds further the losses. In fact, you'd actually be better off building air tanks and pressurizing them and using the air pressure to turn turbines than doing electrolysis, pressurizing hydrogen, and dumping it into a fuel cell. That will give you a loss of only 25% to 50% of the energy that goes in. Sure, it will take up more space, but you won't have hydrogen making the metal brittle after a few years, requiring you to replace the whole system over and over again, so it makes *way* more sense.

When I say that IMO, there is literally no case where hydrogen fuel cells make sense other than space travel, I mean that. It is utterly terrible efficiency-wise, so much so that almost anything is better, including things that are way simpler and cheaper than hydrogen, like a giant air tank and an air turbine.

I would still expect it to be less efficient than electric arc furnaces, but maybe not, so I'll grant you that this might be a very narrow use case, solely because burning the fuel source is actually important for that. :-)

Fuel cell efficiency is at best about 60%. So no, you do NOT have 80% round-trip efficiency, unless you've found a way to violate the first law of thermodynamics.

60% efficiency times 68% is 40.8% efficiency. Yeah, that's slightly better than 34%, but in much the same way that a s**t sandwich is slightly better than s**t. :-)

And this will still be capable of running on natural gas, which probably means it won't be optimal efficiency-wise for either fuel.

The losses from electrolysis alone make it a bad idea, even if the next step were 100% efficient. It just gets worse from there.

Easy for you, a technical person familiar with LLMs and WebAssembly

I'm not talking about how to develop LLM inference servers. You don't have to understand WebAssembly in order to run a WebAssembly program in your browser any more than you have to understand Javascript to run Javascript in your browser. It's *less* technological knowledge than using the Play store. And installing Ollama is no more difficult than installing any other app.

Your difficulty conceptions are simply wrong.

I don't understand your response. Was "life breathed into" the ancient Chinese robotic orchestras and singers, or the Islamic robotic orchestra and mechanical peacocks?

And re: myths, the aforementioned myths literally involved *humans* making the automatons. Ajatasatru for example, the maker of the robots to guard the artifacts of the Buddha, was also famous for using a mechanical war chariot of great complexity with whirling spiked maces, and later one with spinning scythes - not the sort of things you would describe as having "life breathed life into", and actually quite similar to Leonardo Da Vinci's chariot (in some versions he made it/them, in other versions it was a gift from the Indras). As for the robots guarding the Buddha, in one version they're literally powered by water wheels. In another version, Greco-Romans had a caste of robot makers, and to steal the technology, a young Indian man was reincarnated as a Greco-Roman, marries the daughter of a robot-maker, and sews the plans for robots into his thigh, so that when he's murdered by killbots as he tries to flee with the plans, they still make it back to India with his body. Yes, ancient Indian legends literally involved robot assassins.

And as for the robots in the Naravahandatta, they were literally made by a carpenter, and are specifically described as "lifeless wooden beings that mimic life".

Even with Hephestos, a literal god, they're very much not described as merely having life breathed into them - they're literally described as having been crafted (the Greeks were very much into machinery and described it in similar terms), and they behave as if something that were programmed (the Kourai Khryseai are perhaps the most humanlike of Hephaestus's creations, but even they aren't described like you would describe biological beings, they're described for being remarkable for how lifelike they were). Of course it wasn't for-loops and subroutines, people had no conception of such a thing, but his creations behaved in a "programmed" way, not as things with free will.

I don't know why some people are so insistent on imagining that "sci-fi" things have to be recent. They're not. There were literally space operas being written in Roman times. Not scientifically accurate, of course, but sci fi things - including automated things that mimic intelligence - simply is not new.

I'm sorry, but let me repeat: it is an apt description of what it's doing. Nobody says ".... and does so via the exact same mechanism that the human brain uses when paying attention" (which I doubt you could describe anyway, and is in reality very much still a topic of research)

You forgot that this is about gas turbines. They're going to BURN the hydrogen. Divide that 68% number by two, and that's still probably wildly optimistic. More realistic numbers are probably more like 20%.

Not really, no.

If you're rolling it around on wheels, maybe. For a fixed installation, weight has exactly zero relevance. You're putting it on top of a concrete slab on top of dirt. Who cares how much it weighs?

Volumetric density might matter sometimes. Typical density for hydrogen peaks at about 40 kg per cubic meter (assuming Google search isn't lying to me). With a fuel cell, this will maybe give you 1320 kWh. But then you need additional space for the fuel cell itself, plus compressors to compress the hydrogen on the way in.

Batteries give you half the energy density, but that's all you have to have. Electricity in, electricity out.

Which one is more dense depends entirely on A. how quickly you need to store the incoming hydrogen (size/number of compressors) and B. how quickly you need to be able to turn the hydrogen in your tanks into electricity. Because the batteries will be instant. The power is just there. Whereas with fuel cells you need more/bigger fuel cells depending on how high your kW output needs to be. So storing huge amounts of power is more dense with hydrogen if you only need to dribble it out, but massively less dense if you need to dump all of the stored energy in an hour or two.

And realistically, for grid-tied energy storage, that second case is more common than the first. You aren't going to store energy for a year unless you're in Alaska had have all-day twilight for several months. No, you're going to store the energy during the day and use the vast majority of it between the middle of the afternoon to the early evening. It's probably a three or four hour window in which you will be dumping all the energy that you stored, give or take.

But to make matters worse for hydrogen, they're talking about burning it, not using it in a fuel cell. The efficiency there is maybe half the efficiency of a fuel cell. So when used in that way, batteries are more efficient in terms of volumetric density than hydrogen even BEFORE you factor in all the space for the turbines to burn it and turn it into electricity! This is absolutely *insanely* space-inefficient.

Add to that the problem of hydrogen embrittlement, where you have to keep replacing those storage tanks every few years, not to mention the pipes, turbines, etc., and it quickly becomes obvious that this project is a giant money pit in which Southern California will burn dollars and turn them into a negligible amount of temporary power storage.

There's no way in this world that burning hydrogen from electrolysis at somewhere in the neighborhood of 20% round-trip efficiency makes sense. This is quite possibly the single most clueless idea ever to come out of California's government in the history of California's government. The only people this makes sense for are the ones who are bilking the taxpayers by building out this infrastructure. Because it will never be useful. It will always be more efficient to use the incoming energy to charge batteries, or to do something else. Even when you're talking about things like nuclear power and using waste heat to crack water into hydrogen, you'd still be more efficient with any number of other thermoelectric energy capture systems going straight to electricity and storing it in a battery.

Hydrogen is not the answer. Hydrogen is the question. No is the answer. Always. For literally any purpose you could possibly come up with other than fusion.

Fuel cell efficiency sucks, too. Sure, maybe it's half again more efficient or even twice as efficient, but a battery would be more like 5x as efficient.

Oh, and also, the resources needed to do a finetune (to update knowledge) - or heck, even just a LoRa - are vastly less than the resources needed to train a foundation. And in any "AI-crash scenario", renting server time becomes cheap.

Installing Ollama is easy, and now there's even WebAssembly inference servers which load models straight in your browser but run on your computer. Literally just have to browse to the page and click.

RAG does not require any meaningful amount of maintenance.

What if perchance the track should bend?

The Trump administration has been broadly declaring anything related to DEI illegal and putting immense pressure against major organizations with DEI policies, when again, like 99% of policies, including those they've gone after, are like they above.

They literally had to turn down a government grant because they had a DEI policy. They're not lying about the liability risk they faced. And this is what the overwhelming majority (like 99%) of real-world DEI policies look like.