This is perhaps true from the "language model" point of view, but surely from the "knowledge" point of view an LLM is prioritising a few "correct" data sources?
I wonder about this a lot when I ask LLMs niche technical questions. Often there is only one canonical source of truth. Surely it's somehow internally prioritising the official documentation? Or is it querying the documentation in the background and inserting it into the context window?
This is a good article. Not because it's got some crazy insights or radical suggestions - but because it's pragmatic and sensible advice for any project. It definitely resonates with my experience - the biggest risk is just losing focus or losing track of what you're meant to do.
It's refreshingly free of buzzwords and rigid "process" too!
Yeah the hardest thing is to focus intensely and have a strong vision for what exactly the output should be directionally. The second hardest is actually getting the project finished - that requires sustained intense focus.
Theres nothing more to it than that. Frameworks etc blah blah blah. Who cares. Get the work done.
You'd need to write an entire hardware abstraction layer to do anything useful. There's projects that do this for microcontrollers - eg MicroPython and Espruino.
Yes, it would need support from lower level code. But then, so does C -- many things that an OS needs to do, such as installing interrupt handlers, changing the current page table pointer, jumping into a target process already in progress, etc., are not part of the C standard.
If anyone is saying "yeah, but this time will be different", just look at our society now.
Arguably the only jobs which are necessary in society are related to food, heating, shelter and maybe some healthcare. Everything else - what most people are doing - is just feeding the never ending treadmill of consumer desire and bureaucratic expansion. If everyone adjusted their desired living standards and possessions to those of just a few centuries ago, almost all of us wouldn't need to work.
Yet here we are, still on the treadmill! It's pretty clear that making certain types of work no longer needed will just create new demands and wants, and new types of work for us to do. That appears to be human nature.
> "no matter how many books you read about ancient history, you need to have lived there before you can reasonably talk about it"
Every single time I travel somewhere new, whatever research I did, whatever reviews or blogs I read or whatever videos I watched become totally meaningless the moment I get there. Because that sliver of knowledge is simply nothing compared to the reality of the place.
Everything you read is through the interpretation of another person. Certainly someone who read a lot of books about ancient history can talk about it - but let's not pretend they have any idea what it was actually like to live there.
So you're saying that every time we talk about anything we don't have direct experience of (the past, the future, places we haven't been to, abstract concepts, etc.) we are exactly in the same position as LLMs are now- lacking a real world model and therefore unintelligent?
What makes Linux not an option? Is there specific apps you need to use? Or IT policies? Or something else?
The company I work for got bought by a big conglomerate, and I managed to stubbornly hold out using Linux for a really long time. It turns out if your workplace has adopted “Bring your own device” type policies, that often means you can auth with enough services that working on Linux is feasible.
The issue isn’t the technology, it’s that there’s more than one way to do everything and people tend to scratch their itches.
If you started a company today, you can immediately and cheaply hire people or an MSP to manage Windows PCs. I hire entry Windows techs for $70k. M365 E-whatever is $30-60/mo.
Apple fully aligned their products, so the guys running the iPhone fleet can run the Macs. They may need some higher level assistance to setup the configurations. Unless you have a lot of compliance work, enroll in MDM, done.
With Linux, other than Chrome, there’s no standard. You’re gonna need a smart/expensive person to setup things and you’re going to need smart/expensive people to operate. If you have compliance requirements appear, you’ll need to buy RHEL or something and rework stuff, which is more expensive than windows.
It's much harder for non-dev jobs where the management won't let you BYOD for whatever reasons, which could range from IT being too stubborn to allow you to keep company data on your own laptop that's not centrally managed, to everything including licenses for random 3rd party software the company is using being tied to the ActiveDirectory fleet of computers with centralized storage.
This is the reality of IT equipment in big parts of the non-dev world, and you'll have a hard time convincing the IT dept to take on extra hassle just for you to use Linux out of all hundreds of employees who're just fine with Windows.
Linux is good for a lot of things, but the end user has to be comfortable too. They've heard of a Mac, but a Dell with Mint or whatnot is a harder sell.
I almost never use word for exactly that reason. I don’t want to spend half an hour normalising my headings and fonts and margins. I want to focus on content and logical structure.
I much prefer Google Docs over word for this reason too.
I was writing a datasheet really and it’s really surprising how there isn’t ia straightforward solution. Confluence wasn’t expressive enough, while getting Word to apply consistent styles across tables, margins, headings etc is such a pain.
* Everything is being irradiated all the time. Things need to be radiation hardened or shielded.
* Putting even 1kg into space takes vast amounts of energy. A Falcon 9 burns 260 MJ of fuel per kg into LEO. I imagine the embodied energy in the disposable rocket and liquid oxygen make the total number 2-3x that at least.
* Cooling is a nightmare. The side of the satellite in the sun is very hot, while the side facing space is incredibly cold. No fans or heat sinks - all the heat has to be conducted from the electronics and radiated into space.
* Orbit keeping requires continuous effort. You need some sort of hypergolic rocket, which has the nasty effect of coating all your stuff in horrible corrosive chemicals
* You can't fix anything. Even a tiny failure means writing off the entire system.
* Everything has to be able to operate in a vacuum. No electrolytic capacitors for you!
So I guess the question is - why bother? The only benefit I can think of is very short "days" and "nights" - so you don't need as much solar or as big a battery to power the thing. But that benefit is surely outweighed by the fact you have to blast it all into space? Why not just overbuild the solar and batteries on earth?
The main reason is that generating energy in space is very cheap and easy due to how ridiculously effective solar panels are.
Someone mentioned in the comments on a similar article that sun synchronous orbits are a thing. This was a new one to me. Apparently there's a trick that takes advantage of the Earth not being a perfect sphere to cause an orbit to precess at the right rate that it matches the Earth's orbit around the sun. So, you can put a satellite into a low-Earth orbit that has continuous sunlight.
Is this worth all the cost and complexity of lobbing a bunch of data centers into orbit? I have no idea. If electricity costs are what's dominating the datacenter costs that AI companies are currently paying, then I'm willing to at least concede that it might be plausible.
If I were being asked to invest in this scheme, I would want to hear a convincing argument why just deploying more solar panels and batteries on Earth to get cheap power isn't a better solution. But since it's not my money, then if Elon is convinced that this is a great idea then he's welcome to prove that he (or more importantly, the people who work for him) have actually got this figured out.
Let's assume your space solar panel is always in sun - so 8760 kWh per year from 1kWp.
In Spain, 1kWp of solar can expect to generate about 1800 kWh per year. There's a complication because seasonal difference is quite large - if we assume worst case generation (ie what happens in December), we get more like 65% of that, or 1170 kWh per year.
That means we need to overbuild our solar generation by about 7.5x to get the same amount of generation per year. Or 7.5kWp.
We then need some storage, because that generation shuts off at night. In December in Madrid the shortest day is about 9 hours, so we need 15 hours of storage. Assuming a 1kW load, that means 15kWh.
European wholesale solar panels are about €0.1/W - €100/kW. So our 7.5kWp is €750. A conservative estimate for batteries is €100/kWh. So our 15kWh is €1500. There's obviously other costs - inverters etc. But perhaps the total hardware cost is €3k for 1kW of off-grid solar.
A communications satellite like the Eurostar Neo satellite has a payload power of 22 kW and a launch mass of 4,500 kg. Assuming that's a reasonable assumption, that means about 204kg per kW. Current SpaceX launch costs are circa $1500 per kg - but they're targeting $100/kg or lower. That would give a launch cost of between $300k and $20k per kW of satellite power. That doesn't include the actual cost of the satellite itself - just the launch.
I just don't see how it will make sense for a long time. Even if SpaceX manage to drastically lower launch costs. Battery and solar costs have also been plummeting.
Is it reasonable to use Neo as a baseline? Modern Starlink satellites can weigh 800kg, or less than 20% of Neo. I see discussions suggesting they generate ~73kw for that mass. I guess because they aren't trying to blanket an entire continent in signal? Or, why are they so much more efficient than Neo?
Interestingly the idea of doing compute in space isn't a new one, it came up a few years ago pre-ChatGPT amongst people discussing the v2 satellite:
Still, you make good points. Even if you assume much lighter satellites, the GPUs alone are very heavy. 700kg or so for a rack. Just the payload would be as heavy as the entire Starlink satellite.
You can't increase the size of the radiator and reduce the mass of the satellite. How is that supposed to work?
You're also forgetting that Starlink satellites aren't in a sun synchronous orbit which means they have to overbuild the energy generation capacity (low capacity factor) and can simultaneously take advantage of earth's shadow to cool down.
Droplet radiators can theoretically do this. The radiator is made up of extremely fine liquid droplets expelled from what is basically a big space shower head. The droplet cloud has a big surface area so more heat can radiate. The droplets are collected in a sort of drain the other side. The idea has been around for decades but there are lots of practical problems to work out, like minimizing losses due to splashes or droplets heading in the wrong direction (e.g. using ferrofluids and magnetic containment). It's never been worked on seriously because conventional radiators were always enough.
With droplet radiators increasing the effective size means using a bigger head/drain and longer booms to expand the distance between them, so the scaling properties are different to pipe based radiators.
That could be one reason they want to do it. Maybe by using data from Palantir or harvested from Elon's work with DOGE, along with twitter user data and whatever else they can get, they want their AI to be the all-seeing eye of Sauron. (Which isn't too far from what the whole ad-tech industry is about in the first place.) Or they want to make sexually explicit deepfakes of everyone Elon doesn't like. Or they want to flood the internet with AI generated right-wing propaganda.
If one kilogram of stuff consumes just 100Wt, then in one month it consumes about 300 MJ. So as long as things works for a year or more energy cost to put them into orbit becomes irrelevant.
To keep things in orbit ion thrusters work nicely and require just inert gases to keep them functioning. Plus on a low Earth orbit there are suggestions that a ramjet that capture few atoms of atmosphere and accelerates them could work.
Radiative cooling scales by 4th power temperature. So if one can design electronics to run at, say, 100 C, then calling would be much less problematic.
But radiation is the real problem. Dealing with that would require entirely different architecture/design.
Because the permitting process is much easier and there are way, way fewer authorities that can potentially shut you down.
I think this is the entire difference. Space is very, very lightly regulated, especially when it comes to labor, construction and environmental law. You need to be able to launch from somewhere and you need to automate a lot of things. But once you can do this, you escaped all but a few authorities that would hold power over you down on Earth.
No one will be able to complain that your data center is taking their water or making their electricity more expensive, for example.
The satellite is built on Earth, so I’m not sure how it dodges any of those regulations practically. Why not just build a fully autonomous, solar powered datacenter on Earth? I guess in space Elon might think that no one can ban Grok for distributing CSAM?
There’s some truly magical thinking behind the idea that government regulations have somehow made it cheaper to launch a rocket than build a building. Rockets are fantastically expensive even with the major leaps SpaceX made and will be even with Starship. Everything about a space launch is expensive, dangerous, and highly regulated. Your datacenter on Earth can’t go boom.
Truly magical thinking, you say? OK, let's rewind the clock to 2008. In that year two things happened:
- SpaceX launched its first rocket successfully.
- California voted to build high speed rail.
Eighteen years later:
- SpaceX has taken over the space industry with reusable rockets and a global satcom network, which by itself contains more than half of all satellites in orbit.
- Californian HSR has spent over thirteen billion dollars and laid zero miles of track. That's more than 2x the cost of the Starship programme so far.
Building stuff on Earth can be difficult. People live there, they have opinions and power. Their governments can be dysfunctional. Trains are 19th century technology, it should be easier to build a railway than a global satellite network. It may seem truly magical but putting things into orbit can, apparently, be easier.
That’s a strange comparison to make. Those are entirely different sectors and sorts of engineering projects. In this example, also, SpaceX built all of that on Earth.
Why not do the obvious comparison with terrestrial data centers?
Now how about procuring half a gigawatt when nearby residents are annoyed about their heating bills doubling, and are highly motivated to block you? This is already happening in some areas.
From individual POV yes, but already Falcons are not that expensive. In the sense that it is feasible for a relatively unimportant entity to buy their launch services.
"The satellite is built on Earth, so I’m not sure how it dodges any of those regulations practically."
It is easier to shop for jurisdiction when it comes to manufacturing, especially if your design is simple enough - which it has to be in order to run unattended for years. If you outsource the manufacturing to N chosen factories in different locations, you can always respond to local pressure by moving out of that particular country. In effect, you just rent time and services of a factory that can produce tons of other products.
A data center is much more expensive to build and move around. Once you build it in some location, you are committed quite seriously to staying there.
It would make more sense to develop power beaming technology. Use the knowledge from Starlink constellations to beam solar power via microwaves onto the rooftops of data centers
I guess in terms of the relative level of stupidity on display, it would be slightly less stupid to build huge reflectors in space than it is to try to build space datacenters, where the electricity can only power specific pieces of equipment that are virtually impossible to maintain (and are typically obsolete within a few years).
Almost none of the parent’s bullet points are solved by building on the Moon instead of in Earth orbit.
The energy demands of getting to the 240k mile Moon are IMMENSE compared to 100 mile orbit.
Ultimately, when comparing the 3 general locations, Earth is still BY FAR the most hospitable and affordable location until some manufacturing innovations drop costs by orders of magnitude. But those manufacturing improvements have to be made in the same jurisdiction that SpaceXAI is trying to avoid building data centers in.
This whole things screams a solution in search of a problem. We have to solve the traditional data center issues (power supply, temperature, hazard resilience, etc) wherever the data centers are, whether on the ground or in space. None of these are solved for the theoretical space data centers, but they are all already solved for terrestrial data centers.
But none of those are usable, right? It will take decades of work at least to get a commercial grade mining operation going and even then the iron, titanium, aluminum would need to be fashioned...
Ah, I see the idea now. It is to get people to talk about robotics and how robots will be able to do all this on the moon or wherever.
That's a hard problem to solve. Invest enough in solving that problem and you might get the ability to manufacture a radiator out of it, but you're still going to have to transport the majority of your datacenter to the moon. That probably works out more expensive than launching the whole thing to LEO
Sounds more difficult. Not only is the moon further, you also need to use more fuel to land on it and you also have fine, abrasive dust to deal with. There’s no wind of course, but surely material will be stirred up and resettle based on all the landing activity.
And it’s still a vacuum with many of the same cooling issues. I suppose one upside is you could use the moon itself as a heat sink (maybe).
And 2.5s is best case. Signal strength issues, antenna alignment issues, and all sorts of unknown unknowns conspire to make high-integrity/high-throughput digital signal transmissions from a moon-based compute system have a latency much worse than that on average.
Yeah, carrying stuff 380k km and still deploying in vacuum (and super dusty ground) doesn't solve anything but adds cost and overhead. One day maybe, but not these next decades nor probably this century.
Still a vacuum so the same heat dissipation issues, adding to it that the lunar dust makes solar panels less usable, and the lunar surface on the solar side gets really hot.
My neighbour has a motion activated flood light. It's annoying. Not annoying enough to risk a feud by telling them though. It also completely ruins any natural habitat for nocturnal animals.
The whole concept of permanently lighting your garden is crazy! Where do you live that you're so worried about people hiding in your yard? Could you not solve that with cameras and an infra-red floodlight?
Even infrared is weird to me. Insects and other creatures living in the garden have issues with it, while they are important for a healthy environment ...
It mostly leaks and such.
Limescale buildup is also a small issue for their efficiency and more so if they run hot.
If we reduced it to a simple input output calculation that would never be an issue except for some speed of transfer.
I wonder about this a lot when I ask LLMs niche technical questions. Often there is only one canonical source of truth. Surely it's somehow internally prioritising the official documentation? Or is it querying the documentation in the background and inserting it into the context window?
reply