So why don't we do it? FUD.

A consortium of Utah's utilities (UAMPS) literally just pulled out of its commitment to backing NuScale's modular reactor in November 2023. It was a problem of cost, when the construction looked like it was going to become too expensive, at a time when new wind construction is dropping the price of wind power. It basically just couldn't compete on cost, in the specific environment of servicing Utah.

geothermal is probably expensive due to hard rock

I wouldn't sleep on geothermal as a future broad scale solution for dispatchable (that is, generation that can be dialed up and down on demand) electrical power. The oil and gas fracking industry has greatly improved their technology at imaging geological formations and finding places where water can flow and be pumped, in just the past decade. I expect to see over the next decade geothermal reach viability beyond just the places where geothermal heat is close to the surface.

We can’t talk about things like this like they’re free.

Some shifts genuinely are free, though. Wholesale prices for electricity follow a pronounced "duck curve," and drop to near zero (or even negative) in areas where there's a substantial solar base, during the day at certain parts of the year. People will shift their demand for non-time-sensitive consumption (heating, cooling, charging of devices/EVs, batched/scheduled jobs) in response to basic price signals. If a substantial amount of future demand is going to be from data centers performing batched/scheduled jobs, like training AI models or encoding video files, a lot of that demand can be algorithmically shifted.

There are already companies out there intentionally arbitraging the price differences by time of day to invest in large scale storage. That's an expensive activity, that they've determined is worth doing because there's profit to be made at scale.

At household scale, individuals can do that too.

Put another way, we shouldn't talk about current pricing models where every kilowatt hour costs the same as if that arrangement is free.

Plus, the timing of consumption already does naturally tend to follow the timing of solar generation. Most people are more active during the day than at night, and work hours reflect that distribution. Overcapacity in solar can go a long way towards meeting demand when it naturally happens.

Up our storage game, big time

I think this can be expanded out a bit, to the more generalizable case of matching generation to demand. Yes, storage can be a big part of that.

But another solution along the same lines may be demand shifting, which in many ways, relies on storage (charging car batteries, reheating water tanks or even molten salt only when supply is plentiful. And some of that might not be storage, per se, but creating the useful output of something that actually requires a lot of power: timing out industrial processes or data center computational tasks based on the availability of excess electrical power.

Similarly, improvements in transmission across wide geographical areas can better match supply to demand. The energy can still be used in real time, but a robust enough transmission network can get the power from the place that happens to have good generation conditions at that time to the place that actually wants to use that power.

There's a lot of improvement to be made in simply better matching supply and demand. And improvements there might justify intentional overbuilding, where generators know that they'll need to curtail generation during periods where there's more supply than demand.

And with better transmission, then existing nuclear plants might be able to act as dispatchable backup power rather than the primary, and therefore serve a larger market.

When costs are level per kilowatt over lifetime Nuclear is cheaper thanks to economies of scale

Citation needed.

Vogtle added 2000 megawatts of capacity for $35 billion over the past 15 years. That's an up-front capital cost of $17,500 per watt. Even spread over a 75 year expected lifespan, we're talking about $233 per watt per year, of capital costs alone.

Maintenance and operation (and oh, by the way, nuclear is one of the most labor intensive forms of energy generation, so you'll have to look at 75 years of wage increases too) and interest and decommissioning will add to that.

So factoring everything in, estimates are that it will work out to be about $170/MWh, or $0.17 per kwh for generation (before accounting for transmission and reinvestment and profit for the for-profit operators). That's just not cost competitive with anything else on the market.

Economies of scale is basically the opposite of the problem that 21st century nuclear has encountered, which is why the current push is to smaller reactors, not bigger.

There's a place for extending nuclear power plant lifespans as long as they'll go. There's less of a place for building new nuclear.

When they say modules, does that mean mainboards?

They mean each part. Here's their store for individual parts.

This announcement includes a new display, so anyone with the old display can swap out their old one for the new one. People can swap out batteries. Keyboards. Touchpads. It's a modular design so that each module can be swapped out if broken, or if there's been an upgrade the user wants.

I sync if I have a good Internet connection, like from my hotel room or whatever, by VPNing into my home network where my NAS is. There are distributed DNS type solutions for a lot of the big NAS brands, where they'll let you access your data through their service, but I never set that up because I already have a VPN. So my NAS and firewall are configured not to allow outside connections to that device.

But if I haven't synced laptop to NAS yet, then copies exist on both my camera SD cards (redundant double SD card) and my laptop.

3-2-1 backup is important. I've been burned with lost files before, so I now make sure they're available in multiple places.

I also encrypt everything. My laptops can't be unlocked by anyone except myself: Apple Filevault on my Apple laptop, LVM on LUKS on my Linux laptop. If something happens to me, my laptops must be wiped completely to be useable as a used device.

My NAS keeps my backups of all my documents and media (and as a hobbyist photographer, I have over a terabyte of photos and videos I've taken). It's encrypted, but I've written the key down on paper and put it in my physical documents. If something happens to me, someone who goes through my physical documents will have access to my digital files.

I pay a cloud service (Backblaze) for cloud backups. I trust the encryption and key management to not actually give the service provider any access to my files.

The tech companies literally drive around cars looking at what wifi beacons they see, and then store the locations where those BSSIDs were seen.

Oh I actually know this one. Mostly historical accident and path dependence.

In medieval England, kings wanted to make sure that taxes and fines to the crown were properly paid, so they had their own officials in each county, who reported to the King rather than to any local officials. Sheriffs were responsible for tax collection, law enforcement (both arresting people before they could be tried and carrying out the rulings of the court). But they'd have to wait for the king's courts to actually come to town and hold trials and what not, so in the meantime the king's financial interests weren't necessarily aligned with the sheriff's.

So coroners were appointed to watch over county matters and represent the king's financial interests whenever the courts came to town.

When someone was convicted of a capital offense, their property escheated to the crown. That was an important source of revenue for the crown, so coroners would determine whether a dead body was the result of a crime or not, in order to make sure the crown wasn't missing out on some convict money.

Both the Sheriff and coroner positions survived the transition into American governance, but independence and democratic reforms meant that these previously crown-appointed positions needed to become elected positions. Most states kept Sheriffs and Coroners as county officials, and preserved some of their traditional roles and duties. Many coroners offices were renamed to "medical examiner" but basically still preserved the role of keeping stats on deaths. And without appointment by the crown, most states just chose to make these elected positions.

So I gotta ask: what's your plan for addressing your stress levels and your lack of sleep? A prescription from your doctor doesn't squarely address those issues, and they should probably be addressed.

In this house we obey Ohm's law!

I'm still a skeptic of the Nova system into the 4 categories (1: unprocessed or minimally processed, 2: processed ingredients, 3: processed foods, 4: ultra processed foods), because it's simultaneously an oversimplification and a complication. It's an oversimplification because the idea of processing itself is such a broad category of things one can do to food, that it isn't itself all that informative, and it's a complication in that experts struggle to classify certain foods as actual prepared dishes being eaten (homemade or otherwise).

So the line drawing between regular processed food and ultraprocessed is a bit counterintuitive, and a bit inconsistent between studies. Guided by the definitions, experts struggle to place unsweetened yogurt into Nova 1 (minimally processed), 2 (processed culinary ingredients), 3 (processed food) or 4 (ultra processed food). As it turns out, experts aren't very consistent in classifying the foods, which introduces inconsistency in the studies that are performed investigating the differences. Bread, cheese, and pickles in particular are a challenge.

And if the whole premise is that practical nutrition is more than just a list of ingredients, then you have to handle the fact that merely mixing ingredients in your own kitchen might make for a food that's more than a sum of its parts. Adding salt and oil catapults pretty much any dish to category 3, so does that mean my salad becomes a processed food when I season it? Doesn't that still make it different than French fries (category 3 if I make them myself, probably, unless you count refined oil as category 4 ultra processed, at which point my salad should probably be ultra processed too)? At that point, how useful is the category?

So even someone like me, who does believe that nutrition is so much more than linear relationships between ingredients and nutrients, and is wary of global food conglomerates, isn't ready to run into the arms of the Nova system. I see that as a fundamentally flawed solution to what I agree is a problem.

Our data centers and backbone internet/Tier 1 internet providers are basically the best in the world. The US Department of Energy maintains a network with 46 Tbit/s connections between its labs.

I'm mostly going off of this article and a few others I've read. This article notes:

Celebrities have previously won cases over similar-sounding voices in commercials. In 1988, Bette Midler sued Ford for hiring one of her backup singers for an ad and instructing the singer to “sound as much as possible like the Bette Midler record.” Midler had refused to be in the commercial. That same year, Tom Waits sued Frito-Lay for voice misappropriation after the company’s ad agency got someone to imitate Waits for a parody of his song in a Doritos commercial. Both cases, filed in California courts, were decided in the celebrities’ favor. The wins by Midler and Waits “have clear implications for AI voice clones,” says Christian Mammen, a partner at Womble Bond Dickinson who specializes in intellectual property law.

There's some more in there:

To win in these cases, celebrities generally have to prove that their voice or other identifying features are unregistered trademarks and that, by imitating them, consumers could connect them to the product being sold, even if they’re not involved. That means identifying what is “distinctive” about her voice — something that may be easier for a celebrity who played an AI assistant in an Oscar-winning movie.

I think taken with the fact that the CEO made a direct reference to the movie she voiced an AI assistant when announcing the product, that's enough that a normal person would "connect them to the product being sold."

I'm having a hard time seeing why one is fine but the other isn't.

I think the law says that neither is fine, in the context here. The law allows celebrity impersonators to engage in parody and commentary, but not to actually use their impersonation skills to endorse products, engage in fraud, and pretend to be that person being impersonated.

It seems that Apple may be interested in at least requiring authentication that the query comes from an Apple device (or even an Apple-approved API key), which would go a long way in alleviating the security flaw.

I can see some value in the server returning BSSID location data directly (especially with risk of intermittent or slow data connections), but the combination of all the factors seems sloppy.

Yes, but an absence of a proof of the positive is itself not proof of the negative, so if we're in the unprovable unknown, we're still back at the point that you can't prove a negative.

Yes but can you prove by evidence that there is no milk in my cup, if I won't let you look inside?

Apple's got one, so does Google, and Microsoft.

They've got beacon location data, yes, but Apple is the only one that gives up that information without first conforming that the query is coming from someone who sees that BSSID. As OP notes:

In this respect, Apple's Wi-Fi database also differs fundamentally from other Wi-Fi databases, such as the one operated by Google.

If you click through to the paper, it describes 2 approaches for using BSSIDs to identify location:

1. Client submits a query listing each BSSID and its signal strength, and the server calculates position and returns where it believes the query is coming from.
2. Client submits a query listing each BSSID it's interested in, and the server responds with the location of each BSSID so that the client can calculate its own position.

See the problem there? Approach 2 gives more raw information away, by outsourcing the positioning calculation to untrusted clients.

And the paper outlines how Apple goes even further than that:

Apple’s Wi-Fi geolocation API [4] works in the latter manner, but with an added twist: In addition to the geolocations of the BSSIDs the client submits, Apple’s API opportunistically returns the geolocations of up to several hundred more BSSIDs nearby the one requested. These unrequested BSSID geolocations are presumably then cached by the client, which no longer needs to request the locations of the nearby BSSIDs it may soon encounter, e.g., as the user walks down a city street.

It goes on later:

Apple’s WPS API is free and places few restrictions on its use. It requires neither an API key, authentication, nor an Apple device; our measurement software is written in Go and runs on Linux. Moreover, Apple appears to make no attempt to filter physically impossible queries. The BSSIDs submitted to the WPS need not be physically proximate to each other nor to the device submitting the query; Apple’s WPS will respond with geolocations for BSSIDs on two different continents in the same request to a querier on a third.

That's the discussion here. Apple keeps a large database, like many other big tech/mapping firms, but does nothing to keep that database hard for strangers to scrape in bulk.

In contrast, Google uses the first approach and keeps the information a bit more restricted by performing the location calculation at the server:

Han et al. reverse-engineered Google’s WPS’s method of operation [17]. Google’s WPS functions differently than Skyhook’s and Apple’s insofar as Google’s service attempts to geolocate the device submitting the query, providing it with only the device’s computed position given a list of BSSIDs from the client.

So it's possible to run this type of service with this type of database, without sharing BSSID locations with anyone else who asks.

Chess has roughly 10^44 positions. Checkers has roughly 10^20.

That means under that metric, chess is roughly 24 orders of magnitude more complex as checkers.

Tic tac toe has roughly 10^3 positions, or 17 orders of magnitude simpler than checkers.

In other words, the complexity gap between chess and checkers is larger than the gap between checkers and tic tac toe.