I think the $1.5 million per satellite includes the amortized costs from everything else. They’ve launched 10,000 satellites so far, so the other fixed costs are spread around across all the satellites and the service itself. And they have lots of paying customers, including maritime and aviation customers. The rural customers who can be served by the network are already additional revenue, and don’t cost any extra to serve.

Similarly, the development costs of each rocket should be amortized across all the ways the rocket is used, including external paid customers unrelated to Starlink, who just pay for their own payloads to go to space.

Looking it up, SpaceX had $11.4 billion in revenue from Starlink in 2025. As far as I can tell, that segment of their company is profitable, and it’s everything else that is a disaster.

But my point is simple: the useful lifespan of a satellite just changes the amortization calculation. If there are enough customers who will use it, then it can still be cheaper than fiber trenched to a single customer.

the growth itself is hella juiced because the GPUs are only relevant for about 3 years till the new ones are out and make more AI for less power. And they depreciate them over 7 years. More than twice as long as they can or should use the GPUs for.

We don’t actually know this for sure, yet. I had expected the A100 generation (released in 2020) to no longer be profitable to run by now, but the backlog in new data centers being turned on and the high demand from Anthropic and OpenAI still leaves those chips useful for inference. You can rent those 2020 chips out today at some price above what they cost to continue running (300W, so electricity prices of USD $0.20 per kWh would translate into about 6 cents per hour. Prevailing spot prices appear to be about $2/hour right now.

But just because I was wrong on 2020 chips, originally sold for about $15,000 in a low interest rate environment, doesn’t mean that I’m wrong about 2024 chips, the B100s that use 1000W and were sold for $35,000, requiring a ton more specialized cooling, power, and network infrastructure. Or the 2026 R100s that use 2000W, and whose prices I can’t seem to find published anywhere, but were set after the memory companies basically locked in their record breaking prices for their HBM. That’s an unsustainable path and at some point, data centers start struggling to find users willing to pay the bare minimum necessary to continue turning a profit on GPU usage.

I doubt the 2024 chips stay in service to 2031. And I’m really, really skeptical that the 2026 chips stay in service to 2033, especially after NVIDIA switches to yearly release cycles next year.

Fable 5 spawning a herd of Codex 5.6 Sol to write metal shaders.

Fable is Anthropic’s current flagship LLM model (Mythos) with safeguards/restrictions intended to prevent it from writing malware. Version 5 is the latest, released in June, briefly banned by the US Government, and then made available again on July 1.

Codex is OpenAI’s coding-oriented interface for interacting with OpenAI’s models. ChatGPT Sol is the most powerful flagship model, and version 5.4 released on July 9.

Metal is Apple’s programming interface for Apple’s GPUs, and is common to iPhone/iPad/Mac.

Shaders are program functions that set up tasks for a GPU to process visual output, like those that calculate how light interacts with colorful objects of varying reflectivity, or how a house should look when viewed through some fog, etc.

The original post describes what is now a relatively common workflow: tell Anthropic’s most powerful model to manage some cheaper models to do specific tasks and put the output together into something that can be used. As the post shows, it doesn’t always work. And when it fails, it can do so in a very expensive way.

But it isn’t encoding knowledge, it’s encoding word correlations.

I’m saying that humans do this a lot, too. Qualitatively, it’s different, in that this particular batch of frontier LLMs will get things wrong in ways that most human brains wouldn’t, but as a category of error it’s not unique to LLMs.

I know a ton of facts that I learned only through reading, and have no actual firsthand knowledge/experience or ability to test it: Jupiter is larger than Saturn, the atmosphere during the Carboniferous period was high in oxygen, cigarettes cause cancer, Thomas Jefferson owned slaves, the capital of Norway is Oslo. At best, I can cross reference other sources and see that things are consistent with each other. Is my belief in those facts “knowledge,” or is it merely recognizing from my training data that those particular words can validly be presented in that order?

If you ask average people on the street whether FAT32 is a good filesystem for a 64GB removable drive, most of them won’t know, but there are a handful of bullshitters who might confidently parrot back things they can Google but not understand. That’s part of the human condition, too.

I’m by no means an AI booster/enthusiast. I suspect LLMs/transformers are actually a dead end, and expect the upcoming crash to be economically and financially devastating to the tech and financial sectors. But I also have a pretty dim view of human intelligence, too, and see way too many parallels in LLMs as bullshit artists to humans as bullshit artists, too.

It modifies the prompt, aka the input, not the output. It is smuggling 3 bits of secret user/session data in a wrapper that doesn’t look like it contains that data. As the article explains:

So the marker becomes part of the system context sent to the model.

This is a normal timestamp on a prompt:

Today's date is 2026-07-11.

But if your system timezone is a Chinese mainland timezone, it looks like:

Today's date is 2026/07/11.

Then, if your base URL includes a keyword like “deepseek,” it silently replaces the apostrophe from a ' to a ʼ:

Todayʼs date is 2026-07-11.

Or if the base URL has one of the domains on the list, like any .cn domain, it replaces the apostrophe with another apostrophe character:

Today’s date is 2026-07-11.

And if it has both a URL and a keyword on the watchlist, the prompt context includes:

Todayʹs date is 2026-07-11

That’s 3 bits of information: does this system have a mainland Chinese time zone, does the base URL contain a known keyword (associated with Chinese AI competitors) or a known domain (associated with mainland China or its major tech companies). And it sneaks it on by without making it obvious.

That’s steganography.

up to 150kW at full load.

That’s the last generation. They’re moving from Blackwell to Rubin chips now, and the 72-GPU Rubin servers use up to 230 kW.

The typical residential connection in the U.S. has a 24 to 48 kW electrical connection. A block of houses might not have enough power infrastructure to power just one of these server racks.

It’s that they are trying to use statistics to encode entire thought processes into hidden variables from conversation snippets. They want to use statistics to go from many individual interactions to a large model, and then use that model to predict individual interactions again.

Has it been shown that the human brain doesn’t model the world in a similar way, though? A huge portion of human knowledge is both stored and transmitted in the form of language. Lots of human knowledge also follows the garbage in, garbage out theory, where you can have entire areas of knowledge that aren’t actually true but might be internally consistent, at least within certain scopes: conspiracy theories, belief in the supernatural, entire academic disciplines built on a religion or theology that not everyone believes, etc. Or even world building in fiction, the words on a page can be enough to convey ideas such that it “tricks” human brains into filling in the gaps so that they internally see a rich, fleshed out world that is entirely fictional and where specific details might not find strong direct support in the underlying text.

it has no concept of correctness

But statistical weight on what is more or less likely to be correct still makes a difference to objective quality of the outputs. If the model weights are trained on the reality that high quality university texts describe something and reflect some sort of underlying model of what is described using language, then can’t the model itself learn as much as a human could from those words on a page?

All models are wrong, but some can be useful. And different models have different quality in different domains. So although I don’t believe LLMs will overtake the hump of getting ahead of human knowledge, I also don’t believe that any given LLM can be evaluated on quality, and that Facebook’s LLMs are significantly behind other LLMs we see.

And that maybe a huge part of it is its internal process of preparing the model to evaluate the quality of its inputs, such that the output it produces can also score high on quality.

Yeah, but if they spread the cost across many customers, the cost per customer is going to be much smaller, even if it doesn’t last as long before needing a replacement.

If it costs $100,000 to build a fiber line to a single home for 30 years (360 months) that house will need to pay $278/month for 30 years to break even. Throw in interest rates/inflation, and it’ll be more.

But if a satellite that costs $1.5 million to build and launch into orbit can serve even 200 customers for 5 years, that’s only $125/month per customer.

As it stands right now, Starlink serves something like 12 million customers on 10,000 satellites. So that’s an average of 1200 customers served by each satellite, which is what makes $50/month service feasible as a business.

That just means the high up front costs of either trenching fiber or launching satellites need to serve a lot of people to recover that cost. That means the last mile for rural residents tends not to be cost effective for fiber, because there aren’t enough connections served by any given segment.

But making it so any given satellite can serve lots of people in its footprint at any given moment might make it cost effective to serve rural residents.

One common strategy is to run fiber to a specific central location and run point to point microwave antennas to the individual houses/buildings served. That way the fiber itself can carry the traffic of hundreds of users, and each house just needs to have an antenna with line of sight to the place where the fiber is terminated. Rural WISPs have been doing this from before Starlink.

Publication date April 2, based on reporting from Bloomberg on April 1. Anyone have an update 3 months later? I want to see what happened to the remaining half.

Intellectual property is as real as any other property is: it’s all made up by people, but it has the force of social/economic/governmental institutions behind it so it can affect your life.

It’s just that legal concepts tend to attract discussions by people who can’t tell the difference between a descriptive or positive argument (how things are) versus a normative argument (how things should be).

CNBC ran this article earlier this week, too:

Employers who laid off workers citing AI are already starting to regret it

If CNBC of all places is running stories like this, that represents a huge vibe shift.

It’s like responding to your employee losing an arm, ripped off by your tiger, and saying “I’m never going to financially recover from this.”

Basically they’d need about as much in radiator fin surface area as they would have in solar panel area. The ISS has 8 solar array wings, 35m x 12m, that can produce about 30 kW each, or 240 kW total, in sunlight (which is only half the time). The ISS has a complex cooling system, but relies on 4 radiators about 3.1 m x 13.6 m to reject up to 14 kW of heat each (56 kW total) for cooling the solar arrays themselves. The main cooling system uses 6 radiators, each 23.3 m x 3.4 m, to reject 70 kW of heat (from this report it sounds like each radiator may be capable of rejecting more than 1/6 of the heat but that the system as a whole needs to be kept under 70 kW of heat rejection).

So that seems like about 650 square meters of radiators can provide about 120 kW of heat rejection.

Today, a 72-GPU Blackwell server is 130 kW in a single server rack. The next generation rolling out now has 72 Rubin GPUs in a 230 kW server, in a single rack. And that’s not even a “data center.” That’s just a single (albeit very powerful) server. How many can you string together, with networking equipment beaming data connections back down to the ground, before the ratio of solar panels and radiators to the actual ship size becomes unworkable?

That said, it’s technically possible, especially if you can radiate the heat at higher temperatures than the ISS does, as the Stefan-Boltzmann law shows that the hotter the radiator, the more heat it can reject. Just completely infeasible from an engineering and economical standpoint, for any data center that hopes to be relevant in an age of 100+ MW data centers.

“I read your message but have nothing more to add”

ACK

If you’re ok zooming out far enough you can serve a static image of the pale blue dot and a red arrow pointing to it.

Dental 3D printing is a different beast from consumer printers. The dental printers are already certified to actually create regulator approved devices that can go in a patient’s mouth.

All words that mean a direction on a spectrum (or a discrete binary choice) imply that the opposite direction or binary exists. But when you string together a phrase with lots of these binary choices, that only implies that any one word can be replaced by its opposite and still describe something that exists, not necessarily that all of the words can be flipped towards their opposite.

So logically:

1. Calling a dog big implies that there are smaller dogs.
2. Calling a dog mean implies that there are nicer dogs.

So a “big, mean dog” implies that there are small mean dogs and big nice dogs in existence, but does not necessarily imply that there are small nice dogs.

Or, let’s say another example. Imagine a group of three siblings: a tall sister, a short sister, and a tall brother. If I say “hmm let’s call the tall sister over,” I did need to use both words “tall” and “sister,” because each word eliminates one of the choices, but that does not imply that there is a short brother in the mix.

The CBA with the union provides the protections. The bare minimum the state requires for everyone isn’t all that relevant when you have a negotiated contract for much more than that bare minimum, whether it’s for-cause termination protections, mandatory notice and severance pay for layoffs, etc.