I think they get to that a couple of paragraphs later:
> The idea was good, as were many elements of the execution, but there were also problems: some of its statistical methods needed more work, some of its approaches were not optimal, some of its theorizing went too far given the evidence, and so on. Again, we have moved past hallucinations and errors to more subtle, and often human-like, concerns.
> It would require a technological breakthrough that we have not yet imagined.
Maybe, but not necessarily. The necessary breakthrough might have been high-temperature superconducting magnets, in which case not only has it been imagined, but it has already occurred, and we're just waiting for the engineering atop that breakthrough to progress enough to demonstrate a working prototype (the magnets have been demonstrated but a complete reactor using them hasn't yet).
Or it might be that the attempts at building such a prototype don't pan out, and some other breakthrough is indeed needed. It'll probably be a couple of years until we know for sure, but at this point I don't think there's enough data to say one way or the other.
> And already, solar plus storage is cheaper than new nuclear.
It depends how much storage you mean. If you're only worried about sub-24h load-shifting (like, enough to handle a day/night cycle on a sunny day), this is certainly true. If you care about having enough to cover for extended bad weather, or worse yet, for seasonal load-shifting (banking power in the summer to cover the winter), the economics of solar plus storage remain abysmal: the additional batteries you need cost just as much as the ones you needed for daily coverage, but get cycled way less and so are much harder to pay for. If the plan is to use solar and storage for _all generation_, though, that's the number that matters. Comparing LCoE of solar plus daily storage with the LCoE of fixed-firm or on-demand generation is apples-and-oranges.
I think solar plus storage absolutely has the potential to get there, but that too will likely require fundamental breakthroughs (probably in the form of much cheaper storage: perhaps something like Form Energy's iron-air batteries).
One can discuss base load and season shifting all day long. But ultimately fusion will fail for two simple reasons; time and money.
If we started building a fusion commercial scale plant today (ie started by planning, permits, environmental assessments, public consultation, inevitable lawsuits, never mind actual construction and provisioning) it'd come online in what? 10 years? 15 years? 20 years?
Want to deploy more batteries? It can be online in months. And needs no more construction than a warehouse.
Financially fusion requires hundreds of billions, committed now, with revenue (not returns) projected at 10 years away (which will slide.) Whereas solar + storage (lots and lots of storage) requires anything from thousands to billions depending on how much you want to spend. We can start tomorrow, it'll be online in less than 2 years (probably a lot less) and since running costs are basically 0, immediate revenue means immediate returns.
Of course I'm not even allowing for fusion being "10 years" from "ready". It's been 10 years from ready for 50 years. By the time it is ready, much less the time before it comes online, it'll be redundant. And no one will be putting up the cash to build one.
High temperature superconducting magnets are not a panacea for the problems with DT fusion. Those issues follow from limits on power/area at the first wall, and the needed thickness of the first wall; these ensure DT reactors will have low volumetric power density, regardless of the confinement scheme used.
With HTSC magnets, a tokamak much smaller than ITER could be built, but ITER is so horrifically bad that one can be much better than it and still be impractical.
And these are not new issues, they've been known for more than 40 years, but never addressed. From the 1983 Led
> But even though radiation damage rates and heat transfer requirements are much more severe in a fusion reactor, the power density is only one-tenth as large. This is a strong indication that fusion would be substantially more expensive than fission because, to put it simply, greater effort would be required to produce less power.
In terms of cost of materials to build a reactor, sure, that seems right. But most of the cost of fission is dealing with its regulatory burden, and fusion seems on track to largely avoid the worst of that. It seems conceivable that it ends up being cheaper for entirely political/bureaucratic reasons.
Relaxed regulatory burden doesn't seem to be making fission competitive in China; renewables are greatly overwhelming it now, particularly solar.
We might ask why regulations are so putatively damaging to nuclear, when they aren't to civil aviation. One possibility is that aircraft are simply easier to retrofit when design flaws are found. If there's a problem with welding in a nuclear plant (for example) it's extremely difficult to repair. Witness the fiasco of Flamanville 3 in France, the EPR plant that went many times over budget.
What would this imply for fusion? Nothing good. A fusion reactor is very complex, and any design flaw in the hot part will be extremely difficult to fix, as no hands on access will be allowed after the thing has started operation, due to induced radioactivity. This includes design or manufacturing flaws that cause mere operations problems, like leaks in cooling channels, not just flaws that might present public safety risks (if any could exist.) The operator will view a smaller problem that renders their plant unusable nearly as bad as a larger problem that also threatens the public.
I was struck by a recent analysis of deterioration of the tritium breeding blanket that just went ahead and assumed there were no initial cracks in the welded structure more than a certain very small size. Guaranteeing quality of all the welds in a very large complex fusion reactor, an order of magnitude or more larger than a fission reactor of the same power output, sounds like a recipe for extreme cost.
Regulation is not a problem, and even the construction costs are not terrible. We can take the Rooppur NPP as a base, it produces reliable energy at 6-7 cents per kWh. The reason for cost overruns is simply because NPPs are one-off products, the Western countries don't have a pipeline for NPP production.
If I understand correctly, the cost/year of an engineer in India is maybe 1/3rd that in the US, and for general labor the disparity is even larger. So it shouldn't be too surprising NPP construction in India is cheaper than in the US. India doesn't have a large NPP pipeline, they just have cheaper labor.
Yes, but solar power panels are also mostly produced in China, where engineers still get less than 1/3 of the US/Europe salary.
European power plants will be more expensive, but even with the LCOE of 12 (twice that of Rooppur) it's still going to be way cheaper than storage for areas that get cold weather (Midwest, Germany, most of China).
Anything south of California? Yeah, just get solar+wind, no need to bother with nuclear.
As we pointed out, PV is still trouncing nuclear in China. So if the difference is smaller there, it's still in favor of solar.
Storage is another matter here, but even there costs for batteries have simply collapsed. Understand that massive storage is needed even in a nuclear-powered economy. If all the 283 million cars and trucks in the US were replaced with 70 kWh BEVs, the storage would be enough to power the US grid (at its current average consumption) for 40 hours. That's a lot of batteries. So the demand is there to continue to drive them down their experience curves. In China, they're already around $50/kWh for installed grid storage systems (not just cell price).
The final storage problem, the only reed that nuclear can be clinging to at this point, is long term/seasonal storage. That's needed either to smooth wind variability (~ week scale) or to move solar from summer to winter (~6 months). There are at least two different ways this could be solved: hydrogen and heat. As mentioned elsewhere in these threads, the latter is very promising, with capex as little as $1/kWh of storage capacity and a RTE of about 40%. Should that work out anywhere close to that nuclear would be in a hopeless position anywhere in the world, even at very high latitudes.
> As we pointed out, PV is still trouncing nuclear in China. So if the difference is smaller there, it's still in favor of solar.
Sure. Solar is easy to scale when you don't care about reliability, nobody is arguing with that. But it's another issue entirely when you need a stable grid.
I'm not aware of any countries (even tropical ones) that managed anything close to 100% renewables with solar. E.g. Hawaii has to pay for extremely expensive diesel generation even though they have plenty of solar potential.
And nuclear is scalable if you force other sources off the grid in favor of nuclear (and force customers to not use renewables "behind the meter").
In a fair grid, solar and wind get built out, and the residual demand has no baseload component. Unless nuclear is given the right to force other sources off the grid it becomes inappropriate.
In Texas now there is no chance of new nuclear construction. ERCOT is a competitive market and new nuclear simply doesn't make sense.
First, coal has a much larger share of its cost as variable cost which is avoided if you don't run the plant. 40% for coal, only 10% for nuclear. This makes integrated a coal fired plant into a renewable grid easier than a nuclear plant. China is increasingly doing this with its coal plants.
Second, coal is much more forgiving of maintenance sloppiness, and even in the event of catastrophic malfunction the plant remains repairable.
Nuclear has been available in its current (and no longer competitive) state longer than solar/wind have been in their current economic state, so if you look at historical data you might conclude nuclear is better. But that's backward looking and says little about what's better in the future.
You are aware that a nuclear plant tripping offline was part of the cause of ERCOT's last winter cold problem?
The nuclear power plant wear-and-tear is roughly proportional to the number of hours it runs at full power. By not running the plant, you can extend its service life (probably to more than 100 years, with periodic annealing). The main limiting factor is the reactor vessel, its steel walls can only tolerate so much neutron bombardment before becoming too brittle for service.
Nuclear power plants have similar behavior to coal plants in another regard, they take approximately the same time to ramp up/down.
> You are aware that a nuclear plant tripping offline was part of the cause of ERCOT's last winter cold problem?
They just need to build more of them. Problem solved.
Anyway, nuclear power plants went from 0% to 70% generation in France within 20 years in 70-s. We don't see anything like this happening with solar, even in smaller island countries. Solar is successful only when it's backed by fossil fuels and government subsidies to keep that fossil fuel generation running.
Sure, you can save some maintenance cost by operating at low capacity factor. But this is a minor part of the cost of nuclear energy, so you don't save much. Nuclear simply isn't constituted to be useful as a dispatchable source.
The technical ability to ramp up/down is beside the point; it's the financial ability to do so that matters.
What nuclear did in France half a century ago is irrelevant. What matters is if nuclear makes sense today. It doesn't, even if it could be done.
Nuclear can be made flexible, that's my point. It works best as a constant baseload, but it's mostly because the current plants were not designed for dispatchable use (except for some plants in France).
Nuclear plants do not degrade at a constant rate, regardless of their power. By idling the plant, you extend its service life, essentially amortizing the capital cost over a longer period of time. And the capital cost is the main driver in the cost of the nuclear power, as you're pointing out yourself.
You ignore the counterargument I already gave you to what you're saying there.
Nuclear can be made technically flexible. It can't be made economically flexible. The large fixed costs prevent the technical ability you are describing there from being useful. It's pyrrhic engineering, straining to achieve an outcome that's useless. Even France depends largely on the rest of Europe to deal with variations in demand rather than spooling their power plants up and down.
I think I replied to your counter-argument, but I think I did not explain my argument properly.
In the case of nuclear power plants, the expenses are front-loaded in the construction (and the future major maintenance, like reactor vessel annealing). The _running_ expenses are trivial by comparison. So a nuclear power plant saves a much smaller percentage of its cost on a per-month basis when it's not running.
Honestly, I looked at nuclear energy in a lot of details. It absolutely is a viable and economic path forward, but it stymied by the lack of political will. Nuclear projects take at least 8-10 years to complete, so politicians are less interested in pushing them. And commercial companies are hesitant to invest with such long repayment periods.
> The reason for cost overruns is simply because NPPs are one-off products
But there's no fundamental reason they _have_ to be one-off products. They just historically have been for at least partly regulatorily motivated reasons: because each reactor's approval process starts afresh (or rather, did until quite-recent NRC reforms), there's little advantage in reuse, and because many compliance costs are both high and fixed, there's an incentive to build fewer huge reactors rather than more small ones, which makes factory construction difficult to achieve and economies of scale hard to realize.
Regulatory costs and waste disposal are not significance cost centers for nuclear, at least as far as I can tell from any cost breakdowns.
One doesn't need super high quality welding and concrete pours becuase of regulations as much as the basic desire to have a properly engineered solution that lasts long enough to avoid costly repairs.
Take for example this recent analysis on how to make the AP1000 competitive:
There are no regulatory changes proposed because nobody has thought of a way that regulations are the cost drivers. Yet there's still a path to competitive energy costs by focusing hard on construction costs.
Similarly, reactors under completely different regimes such as the EPR are still facing exactly the same construction cost overruns as in the rest of the developed world.
If regulations are a cost driver, let's hear how to change them in a way that drives down build cost, and by how much. Let's say we get rid of ALARA and jack up acceptable radiation levels to the earliest ones established. What would that do the cost? I have a feeling not much at all, but would like to see a serious proposal.
One approach would be to reduce the size of the containment building by greatly reducing the volume of steam it must hold. This would be done by attaching Filtered Containment Venting Systems (FCVS) that strip most of the radioactive elements from the vented steam in case of a large accident.
The containment building is a significant cost driver, costing about as much as the nuclear island inside of it.
If such a system had been attached to the reactors that melted down at Fukushima exposure could have been reduced by maybe two orders of magnitude. And if the worst case exposure is that low, perhaps much more frequent meltdowns could be tolerated, allowing relaxation of paperwork requirements elsewhere.
Oh for sure, I'm not claiming that CFS (or Tokamak Energy or Type One or whoever else) will for sure succeed, or if they do, that they've already solved all the problems that will need solving to do so. My only assertion/prediction is that I think if they end up succeeding, when future historians look back and write the history of this energy revolution or whatnot, HTSC magnets will turn out to have been the key breakthrough that made it possible.
> If the plan is to use solar and storage for _all generation_, though, that's the number that matters.
And that's the problem with these Internet discussions: that's almost never the plan, but commenters trying to make solar look bad assume it is (to your credit, you made it explicit; many commenters treat it as an unspoken assumption).
In real life, solar and batteries is almost always combined with other forms of generation (and other forms of storage like pumped hydro), in large part due to being added to an already existing large-scale grid. The numbers that matter are for a combination of existing generation (thermal power plants, large-scale hydro, etc) with solar plus storage. Adding batteries for just a few hours of solar power is enough to mitigate the most negative consequences of adding solar to the mix (non-peaking thermal power plants do not like being cycled too fast, but solar has a fast reduction of generation when the sun goes down; batteries can smooth that curve by releasing power they stored during the mid-day peak).
In the end we're still making steam and running a turbine. Just the steam turbine part of the power plant has a hard time competing with solar in sunny locations.
> What is the alternative
Other markets in the US are generally energy + capacity markets -- you get paid both for what you actually provide and for your ability to provide a certain level of power, whereas Texas is an energy-only market (EOM). It needn't be the case that that if you don't do an EOM, you have to have a monopoly.
Definitely you could operate on a capacity model instead of generation. There are a lot of levers. My issue is mostly around how much uninformed hate a “free market” energy system gets.
The whole thing that differentiates this company from the dozen other seemingly-interchangeable new-space entrants is the novel technology they've developed to facilitate reuse. Even if it were the case that there isn't a market for five tons to LEO (and to be clear, Rocket Lab seems to be doing decent business launching a lot less) and all this was was a technology demonstrator, why would you build a technology demonstrator that doesn't show off the thing that makes your company interesting?
> Amazon Kuiper is positioning to compete with SpaceX’s Starlink broadband constellation but it would not rule out seeking launch services from its competitor given the tight deadline, Limp said. “We are open to talking to SpaceX. You’d be crazy not to, given their track record.”
> The Falcon 9 [22.8 tons to LEO], however, is not as large as Amazon would like it to be in order to get maximum bang for its launch buck, as Kuiper satellites are larger than Starlink’s.
> “I would say Falcon 9 is probably at the low end of the capacity that we need,” Limp said. Perhaps a better option would be Falcon Heavy or the much larger Starship, which is still in development. As Starship transitions to production readiness, “that becomes a very viable candidate for us as well.”
Maybe, but they also say Starship would be the best option. There is no way Nova or any other rocket could compete here. Nova would only be good relative to Starship for launching individual smaller payloads into specific orbits, so not for satellite constellations.
I'm not sure the goal of this competition, in and of itself, is AGI. They point to current LLMs emerging from transformers, which in turn emerged from a general basket of building blocks from machine-translation research (attention, etc.). It seems like the suggestion is that to get from where we are now to AGI, some fundamental building blocks are missing, and this is an attempt to spur the development of some of those building blocks, but by analogy with LLMs, the goal here is to come up with a new thing like "attention," not a new thing like GPT4.
Intent on whose part, though? Like, supposing in arguendo that the company's goal was to make the voice sound indistinguishable from SJ's in Her, but they wanted to maintain plausible deniability, so instead cast as wide a net as possible during auditions, happened upon an actor who they thought already sounded indistinguishable from SJ without special instruction, and cast that person solely for that reason. That seems as morally dubious to me as achieving the same deliberate outcome by instruction to the performer.
> happened upon an actor who they thought already sounded indistinguishable from SJ without special instruction, and cast that person solely for that reason
so who was doing the selecting, and were they instructed to perform their selection this way? If there was a law suit, discovery would reveal emails or any communique that would be evidence of this.
If, for some reason, there is _zero_ evidence that this was chosen as a criteria, then it's pretty hard pressed to prove the intent.
I think this is overthinking it. ChatGPT is billed as a general-purpose question-answerer, as are its competitors. A regular user shouldn't have to care how it works, or know anything about context or temperature or whatever. They ask a question, it answers and appears to have given a plausible answer, but doesn't actually do the task that was asked for and that it appears to do. What the technical reasons are that it can't do the thing are interesting, but not the point.
But it it like asking a person for them to generate the same thing, but when the start to list off their answers stopping them by throwing up your hand after their first response, writing that down, and then going back in time and asking that person to do the same thing, and stopping them again, and repeat- and then being surprised after 1000 times that you didn't get a reasonable distribution.
Meanwhile if you let the person actually continue, you may actually get 'Left..... Left Right Left Left... etc'.
If You asked me to pick a random number between one and six and ignore all previous attempts, I would roll a die and you would get a uniform distribution (or at least not 99% the same number).
If you are saying that this thing can't generate random numbers on the first try then it can't generate random numbers. Which makes sense. Computers have a really hard time with random, and that's why every computer science course makes it clear that we're doing pseudo random most of the time.
>If You asked me to pick a random number between one and six and ignore all previous attempts, I would roll a die and you would get a uniform distribution
I believe what GP is getting at is that if you didn't have a die, and you truly ignored all your previous attempts to the point of genuinely forgetting that the question had been asked, then your answer would likely be the same every time. Imagine asking a person with severe Alzheimer's to pick a number, then asking again a few minutes later. You'd probably get the same answer.
Yeah, I get what they are saying and there's no reason to believe that.
The die is an analogy for our decision making. They are implicitly claiming that randomness must come in an order and that simply isn't how randomness works or even halfway decent pseudo randomness.
Any system whether it be a die, a person, or an llm doesn't have to know about its previous random choices to make random choices that follow some distribution going forward presuming it's actually capable of randomness.
I'm just impressed that it answered "left" rather than outputting python code that I could run which would sample the list ["left", "right"] with an 80% bias.
It wouldn't be 0 dollars, though; the majority of their users are apparently outside the US. So the question is: how does the amount you could get by selling a US-inclusive Tiktok compare the potential future earnings of a US-free Tiktok? If the market prices it accurately, you'd sort of expect the former to be higher (a US-inclusive version seems obviously more valuable), but maybe they think the market undervalues them, or maybe prospective buyers would smell blood in the water because of the deadline and try to low-ball, etc.
> The idea was good, as were many elements of the execution, but there were also problems: some of its statistical methods needed more work, some of its approaches were not optimal, some of its theorizing went too far given the evidence, and so on. Again, we have moved past hallucinations and errors to more subtle, and often human-like, concerns.