illiarian

> Right, except at 5x the cost.

Source?

> if grid scale storage were prohibitively expensive or impossible.

It's currently both [1].

BTW. Case in point: https://twitter.com/energybants/status/1647799729734971396

---

[1] Edit: This is a nice quick overview from a year ago https://www.nsenergybusiness.com/features/energy-storage-ana...

"Cheapest" option is pumped hydro storage. To cover Europe's need you only need to build 70 1.5 GW hydroelectric stations at a cost of $92 billion (in reality much higher) while greatly damaging ecology in large areas.

The most expensive option is Li-ion batteries, clocking in at a measely 1.1 trillion dollars.

Cheap and feasible.

> a signal that they'd misjudged and significantly overbuilt capacity, not that it's a good idea.

Unlike, say, renewables, that have overbuilding as a requirement I guess: https://twitter.com/energybants/status/1647799729734971396

What this means is that they built enough capacity to counter most requirements for power, and hand handle elastic demand without the need to overbuild or extensive expensive storage (something also absolutely required for renewables).

> tl;dr - as I said before, you technically can do it but it's absurdly expensive and just not worth it.

And, of course, have sources for this?

> It's still horribly expensive and slow to build.

Many of those delays are political rather than technical.

On top of that, imagine if instead of deriding, denigrating and spreading FUD about nuclear energy for the past 20-30 years we would instead invest in it and improve it.

> the "black swan" disaster,

Literally yesterday there was a black swan for renewables in Germany. They shut down the last reactors and... https://twitter.com/energybants/status/1647799729734971396

--- start quote ---

Germany's 66.5 GW of installed wind is only producing as much as the 3 reactors that turned off last night used to produce.

400% expansion of wind would only provide half present need

--- end quote ---

But of course no renewables proponent will ever mention this as a black swan event. Because "sometime 40 years from now, maybe, we will have overbuilt all the renewables and energy storage we need"

> While most nuclear power plants in operation as of early 2000's were already designed with strong load following capabilities, they might have not been used as such

Again. Very, very VERY slowly:

--- start quote ---

Nuclear power plants in France and in Germany operate in load-following mode

--- end quote ---

And these countries have more power plants than the rest of Europe.

If you bothered to read the report, and not just wikipedia, you'd see graphs for plants built as early as 1978 happily operating in load-following mode. E.g. Unterweser, commissioned in 1978, operated between 700 MW and 1400 MW per day in load-following mode. Same for Grohnde (commissioned in 1984). Etc.

> Find a place that has willing neighbors

You mean Germany right now after it stupidly shut down its last nuclear reactors, and is dependent on coal, and imports from France (75% of whose electricity comes from nuclear).

> You can technically run them in load following mode

The quote literally is: "Nuclear power plants in France and in Germany operate in load-following mode".

Edit: the requirements are literally "For example, according to the current version of the European Utilities Requirements (EUR) the NPP must at least be capable of daily load cycling operation between 50% and 100 % of its rated power Pr, with a rate of change of electric output of 3-5% of Pr per minute."

> but the cost is even more insane than using it as baseload.

And you have sources for that "insane cost"?

> Secondly, it's not a responsive source of electricity - so you can't modulate the output easily.

Yes. Yes, you can.

--- start quote ---

Modern nuclear plans with light water reactors are designed to have strong manoeuvring capabilities. Nuclear power plants in France and in Germany [ report from 2011 ] operate in load-following mode, i.e. they participate in the primary and secondary frequency control, and some units follow a variable load programme with one or two large power changes per day.

Most of the modern designs implement even higher manoeuvrability capabilities, with the possibility of planned and unplanned load-following in a wide power range and with ramps of 5% Pr [rated power] per minute.

--- end quote ---

https://www.oecd-nea.org/upload/docs/application/pdf/2021-12...

> the more modern reactors have been prone to massive delays with an

This is mostly a political issue. China and South Korea are bringing reactors online faster.

> Fourthly, as we found last year, during heat waves rivers can run dry causing a shortage of water to cool the plants forcing them to go offline.

Again, that was a combination of things, including politically-motivated underinvestment in maintenance of nuclear power plants

Hitory, lax laws (esp. around consumer protection) etc.

On top of that the US is a rather homogenous market with a single language vs. 27 conuntries with 27 different languages (in reality more) and quite a difference in local laws (even if all are compliant with EU-wide laws). IIRC there's also an expectation in Europe that companies should actually turn a profit at one point (though with influx of American money this is starting to change, too).

You still get ASML and Infineon, SAP, Adyen, Amadeus etc.

> That’s the reason that it has such a thriving tech ecosystem.

By "market leaders" you mean the very many American companies fuelled by unlimited investor money that lose billions of dollars a year with no chance of ever turning a profit?

> Google is much more diversified (total revenue is north of 250 billion). They have cloud, youtube, docs, google play, all the alphabet experiments etc.

Do read their 10-K: https://abc.xyz/investor/static/pdf/2022Q4_alphabet_earnings...

The only thing that realistically brings in money is ads. All other things are just subsidised by them.

For example Google Cloud (which includes both GCP and workspace offerings aka docs)? Well, it lost 6 billion dollars.

> cause they really need businesses that produce a few hundred billion to keep all their free stuff

The only such business they have is ads.

> It actually seems pretty amazing that what HN thinks is Google's failed side hustle is comparable to the entire business of what would have been considered tech giants a few years ago.

Because it all pales in comparison to what actually powers Google: ads. To the point that there are persistent and growing rumours that GCP may be if not on the chopping block, but greatly de-prioritised (it's part of the 26-billion Google Cloud).

It's also weird to me that people always look at revenue only. As if net income doesn't matter.

For example [1], non-ad business may have brought in 60 billion in revenue, but what it is I see: "The increase in other cost of revenues from 2021 to 2022 was primarily due to increases in data center costs and other operations costs as well as hardware costs." And that cost is 77 billion.

Guess what Google's the non-ad businesses are: "Google other revenues increased ... from 2021 to 2022 primarily driven by growth in YouTube non- advertising and hardware revenues", "Google Cloud's infrastructure and platform services were the largest drivers of growth in Google Cloud Platform."

So a large chunk of that revenue is eaten by the costs of generating that revenue. Google cloud lost 3 billion, other bets lost 6 billion etc.

Meanwhile all other costs of revenue are a 49 billion, compared to over 200 billion in revenue from ads.

So how do those businesses inderectly depend on ads? Ads business subsidises them. Despite the size of those other businesses very few, if any, are successful (as in: actually earn money).

[1] All quotes and numbers from Google's own 10-K: https://abc.xyz/investor/static/pdf/20230203_alphabet_10K.pd...

> Point to the Russian documents with the same clarity and direction that one can, for example, point to the PNAC policy statements [1], which resulted in Iraq losing 5% of its population.

Current Russian regime knows that it cannot clearly explain the invasion of Ukraine. And yet, they invaded, occupied territories, quickly annexed them etc.

Something something "actions speak louder than words".

Net profit margins for full-service restaurants are 2-3%. 2018 was an exceptional year with margins rising to 6%, but I doubt this continued.

https://www.forbes.com/sites/sageworks/2018/01/26/restaurant...

Australia cites 4.2% https://treasury.gov.au/sites/default/files/2019-03/360985-R...

etc.

All the modern features that you see on the web have gone through the flag method at one point.

In JPEG XL case it failed not because it was behind a flag, and the failure of one feature doesn't mean that the entire system isn't working.

> Sounds like the flag method is a failure, and just releasing it actually works.

No, "just releasing" doesn't work for the reasons I outlined above.

> Sure you can - you just do it. You just make the necessary fix.

No, you can't. Not simply.

> But you do it quickly, within a short time of launching it. Anyone who was able to make use of it within a couple months, will be able to adapt and fix things.

That's not how the web works. "People who make use of it in a few months" could easily apps and sites serving millions of people. This is a process that applies not only to image codecs, but to, say hardware APIs and interfaces, network protocols etc.

> But there's no reason to take so long.

There are very many reasons from technical to political to anything in between (and sideways, too).

> How are you supposed to get feedback for a feature people can't actually use?

The moment it's out there (that is, not behind a flag), you cannot fix it, update it etc. because people will be relying on the observed behaviour.

The hard part is advertising the feature (at the very least to developers) because there are just too many featurs in modern browsers. Edit: for example, Chrome has Origin Trials https://developer.chrome.com/origintrials/#/trials/active for experimental features.

> https://caniuse.com/jpegxl

This is kinda how standards are supposed to work: at least two independednt implementations [1], released behind a flag to gather developer and user feedback. However it seems that in the end it went nowhere.

[1] See Implementation experience: https://www.w3.org/2021/Process-20211102/#implementation-exp... Among other considerations: "are there independent interoperable implementations of the current specification?"

Then it's internal politics at play. Or perhaps image codecs are not as important to Google.

E.g. Android TV now requires hardware support for AV1 [1] and Google threatened Roku that Rouku would lose Youtube support if it didn't support AV1 in hardware [2]

[1] https://www.protocol.com/bulletins/av1-android-14-requiremen...

[2] https://www.protocol.com/youtube-tv-roku-issues

They have no reservations when its their own formats and they force others to implement them on pain of removing essential apps like Youtube.

> ultimately going to the airport as rent, not to the CIBO food vendor as profit.

In food services in general there's this weird thing that the ultimate price on the product doesn't matter. The food-selling place will have paper-thin margins regardless. It will either be rent, or equipment and cutlery, or the need for highly trained and specialised staff, or... (and often all of those, and more, coupled together).

There's another batch of emails from those times (can't find the link, but they will definitely appear again) where Gates is fuming that Apple has iPods, and none of their partners that have privileged access to Microsoft (like Creative) can come up with anything even closely resembling it. A few years later MS came out with Zune (so they had to do this themselves), but yeah, it was already too late.

--- start quote ---

On January 24, 2017, Khronos hosted an IP-free meeting dedicated to discussion of "WebGL Next" ideas, collided with WebGL working group meeting in Vancouver.[8] Google team presented the NXT prototype implementing a new API that could run in Chromium with OpenGL, or standalone with OpenGL and Metal. NXT borrowed concepts from all of Vulkan, Direct3D 12, and Metal native APIs. Apple and Mozilla representatives also showed their prototypes built on Safari and Servo correspondingly, both of which closely replicated the Metal API

--- end quote ---

(Note: Unlike Chrome, Apple and Mozilla followed the actual standards route: two independent implementations)

--- start quote ---

On February 7, 2017, Apple's WebKit team proposed the creation of the W3C community group to design the API. At the same time they announced a technical proof of concept and proposal under the name "WebGPU", based on concepts in Apple's Metal.

--- end quote ---

And so on https://en.wikipedia.org/wiki/WebGPU

Apple is literally at the forefront of this entire initiative with Mozilla close second. Chrome grudgingly agreed to join the others a full year later.

> Neither Firefox nor Safari are first,

WebGPU is literally built in top of work done by Safari and Firefox.

What I don't like about these discussions is how they completely skip the pragmatic part.

That's why "expression-oriented" Erlang (and Elixir) simply has the single-branch if while the academics still debate whether or not it should, should be optional etc.

You need to build your network on Twitter, too. And it's much easier because you can more easily search for topics and people.

I have no idea what algorithm throws at me because I only use "for you"

> It is organic interaction, which you get at the Fediverse. Much more so than on Twitter

Do you though? If you're unlucky with your instance of choice, there will be as much "organic interaction" as in an abandoned mining town.

Coupled with aversion to search this makes Mastodon "organic" only on large instances and/or only if you specifically know which people on which instances to follow.

> and combine previously unrelated relationships in reasonable and surprising ways.

We've yet to see those surprisng ways despite all the claims.

Note: what they do already is amazing and surprising in itself (such as "write me a song about quantum physics suitable for a 5 year old"). It's still very much shy of "hey there's this new thing previously unthought of".

> You've practically restated the paper's findings! :D The LLM knew nothing about othello; it wasn't shown any rules to be recombined.

Literal quote from the paper:

"As a first step, we train a language model (a GPT variant we call Othello-GPT) to extend partial game transcripts (a list of moves made by players) with legal moves."

And then:

"Nonetheless, our model is able to generate legal Othello moves with high accuracy".

So:

- it knows about the game because it was literally shown the game with only the legal moves

- it doesn't produce legal moves all the time (even though it does so with high accuracy)

That's why I say "the work done is beyond any shadow of the doubt brilliant". Because this is a definite leap forward from the status quo. However, it doesn't imply that the models can invent/predict/come up with novel ways of doing something. This is still strictly within the realm of "given existing data, give back a statistically relevant response".

Could it actually invent Reversi/Othello had it not known about it beforehand?

> Now imagine which percent of restaurants would fail if we also included failures to comply in other subjects?

And your point to all this is?

> No, they will most likely never be inspected because there are far too many businesses to control and thus will never implement or fix their practices.

Funny how it's not far too many for the heath inspectors, and tax agencies, and ...

> They have just as many rights to do it as large companies

Exactly: zero. Edit: that is, zero right to collect any personal data beyond what they need for the service. And definitely no right to siphon and sell it to others with reckless abandon.