The compute cost for DeepSeek V3 was $5.6M. Training costs for traditional US frontier labs model is about $70 to $200M+, with estimates of up to $1B+ to train the next generation models. It's expensive, but it's not even in the ballpark of "There is no viable economic model to do so." Google revenue in Q2 was $1.32 billion *per day*. $1B is trivially less than a day; $5.6M is just 6 minutes of revenue.

The cost problem is in the compute buildout and the subsidized compute to capture users.

> It remains to be seen if even using their models can be profitable,

I presume we can stipulate there exists some tasks that are cheaper to do using an LLM than doing it "by hand". As a such, using an LLM model can be profitable. To me, it is also obvious that the business of serving an LLM model can be profitable. There's large benefits to being able to get tokens quickly, and you can expect people to pay for that, even if we presume they could run the model locally with no maintenance overhead. There's even cases where using the cloud is cheaper than you can serve locally at all - I saw one guy that measured his added cost of electricity when his computer ran a specific LLM, and it was more expensive to run it locally than to buy tokens from the cheapest inference provider for the same model, presumably because they had better hardware and cheaper electricity.

I love a term a friend of mine used to use when talking about AI: "Narrowly superhuman" (before "superintelligence" got to be a thing).

AI is narrowly superhuman. It is able to do specific things that used to be defined as a "hallmark of intelligence" much better than a human. The easiest one? Summing numbers. It's both faster and more precise than a human.

AI gets more and more such areas it can do better than a human.

If we define "Superintelligent" to be "A computer system able to do absolutely everything we can somehow call intelligence better than any human", we will *never* have superintelligence. The things we define as intelligence will be in conflict; we'll land at the system being better at one at the cost of another.

I'm in favor with being careful about this development. But "prohibiting the development of superintelligence" isn't an option. It's already here.

Dividing the yearly US spend on Internet advertising by the number of people in the US that has Internet access gives $900-$970 spend per person per year, or $75-$81 per month.

Please choose the option you want:

1. I want to pay about $75 per month per member in my family into the ad auctions to not see any ads. Ie, keep the infrastructure, but let you opt to pay your share through money rather than attention.
2. I want to prohibit ads, and accept that everybody will have to pay for the sites they want access to, that there will be much fewer services, that there will be much higher transaction costs, and that the poor will have much less ability to access services. (Microtransactions won't solve this, due to the need for trust evaluation of same.)
3. I want some combination of 1 and 2, where some sides of ads are prohibited so it is possible to buy my way out of them at a lower price,. I accept that this will result in the loss of some services and lower access to others for the poor.
4. I want legislation that makes Internet ad blockers work more consistently. I accept that this just shifts the attention cost to other people, and have me getting the benefit without paying the cost.
5. I want to remove Internet ads and subsidize the sites through taxes.
6. I want to increase the requirements of Trust and Safety checks and format restrictions for ads, potentially accepting higher overhead (in what goes to content vs infrastructure) for less annoying/bad ads. I accept that doing this creates significant advantages of scale for ad networks.
7. Other, please describe, including where costs gets shifted with your proposal.

The above are my best attempt at listing possible options and their correct caveats. In most situations, I find ads annoying (who doesn't). Unfortunately, they are the main money highway of Internet content.

DISCLOSURE: I used to work for Google, where I rang infrastructure for stopping bad ads/advertisers and for removing fake clicks. I've also been interested at the use of sub-cent microtransactions to pay for content since the 1990s, though I've become disillusioned as I've learned more about the attention economics involved. Approving individual microtransactions has a higher attention cost than dealing with the ads, and bulking them per-site (through micro-subscriptions) will limit the number of sites that can have content.

OK, which particular aspect of the LLM learning do you think doesn't exist in the human brain? Backprop exists (and the other training methods the brain uses has similar effects), embeddings more or less exists and seems to be very similar to the LLM embeddings (shown by brain activation in various areas linearly matching with embeddings) - see e.g. High-level visual representations in the human brain are aligned with large language models (Adrien Doerig et al, 2025), and speech/text generation by humans has similar error patterns as for LLMs.

They obviously don't learn in the EXACT same way, but there's also clearly a lot of bits that are the same, and I'm fairly sure nobody knows how much is similar yet, particularly not weighted by importance.
 

I was hoping to find out whether this is a viable business model for new startups, but they only ruled that it took too long for Musk to bring the lawsuit.

Got of those, too (the early USB 1 ones, with the exposed ATAPI connector. I ended up buying Iomega's Firewire expansion that attaches on the back of the slim USB and latches on that ATAPI connector, as Firewire 400 had much better bandwidth than USB 1, provided enough power and thus required only a single cable, and I had a cheap Firewire 400 adapter laying around from some video project (funily: the Firewire 400 card was a free bundle bundled with some crappy movie software that was selling poorly and was on heavy sale at the shop I bought it from. Threw the useless CD, kept the Firewire card).

Mine still works too. The most difficult was trying to find the barrel power plug (since back in the days I was mostly using the Firewire attachment and because Firewire provides enough power, I wasn't using the barrel jack much. Nowadays most of my machine are USB only.

Yup. The slim USB were also a good solution to carry data around.
Bring the slim USB and the cables at the university, download shit with the fast bandwidth, then bring the drive back home, plug into the Firewire attachment and load it onto the computer.
Later the university aquired computers (from Dell) that came with ZIP IDE drive built in, so I only carried the Zip250 disks and kept the drive permanently plugged into the Firewire attachement. And almost lost the power barrel adapter as mentioned above.

Yes, that ass-backward approach came in my mind.
Your bank notes are too easy to copy now that color photocopiers and color laser printers are a thing?
- Rest of the world: make better banknotes (see swiss money, euros, etc.)
- USA: make bank note detection software mandatory on each piece of tech (HP and other US manufacturers have a boner at the thoughts of the sudden illegalness of cheaper competitors from countries without that function) and also mandate yellow dot tracking (now in addition the police-state is having a boner, too) (*).
- Rest of the world: why the hell is my color cartridge constantly empty on yellow and why is this preventing my to print even black and white?

Same here:
USA: has a problem of violence, bonkers level of gun proliferation, on tops of tons of ways to make life shitty for everyone (lack of proper health care, social welfare, etc.)
also the USA: lets add "gun detectors" to 3D printers so nobody prints a gun without a serial number. Surely that's the best solution to address all of the above, right?

Trouble is that this time, most 3D manufacturers ARE NOT in the USA.
Most of them are in China, and the US is only a fraction of their exports, and the required function requires magnitude more compute power to implement than the tiny micro-controller that is usually found in those printers and implementing would require massively driving up the cost of the printer.
Chance are high that the manufacturer will just say f-u, and merely just stop selling complete pre-assembled kit to the USA, only stuff that can circumvent the restrictions (e.g., kits with only motor and drivers that require adding a sold-separately microcontroller).

---

(*): fun fact: on some printers (E.g. with very low memory) those "functionnalities" were implemented in the drivers instead.
My ancient HP color lasterjet works this way. There are no yellow dot when I print from CUPS.

It's entirely possible that the "gun detection" is going to be the same: crappy buggy detection +additional privacy invading tracking implemented into the management software shipped next to the 3D printer as the MCU cannot handle that. Circumventable by downloading Octoprint from some european server and running that on a Pi to manage the printers.

Before emissions controls, they'd try to disperse soot and ash over wide area to minimize the impact to individuals, but natural gas burns a lot cleaner than coal, and doesn't really produce soot or ash, so it's not really a concern.

The only thing that article says about it is that the local government did a study and concluded that it wasn't an environmental hazard.

Think about it this way: why would a gas turbine mounted on a trailer emit more pollution than a fixed facility? It's burning the same fuel, with the same emissions controls.

Ok, but we should still be able to add that to the grid. In reality, it is much easier and cheaper to build a 1GW power plant than it is to build a 1GW datacenter.

2GW is basically nothing. The entire grid is something like 1,200GW so you'd need to add a fraction of a percent to cover a very large data center. The fact that people claim this is beyond our capability is preposterous.

It uses exactly the amount of power I think it does, because I can read the specs and they say how much power these things use, which is not very much.

See, this is the kind of retarded bullshit you idiots say, and then you expect me to take you seriously.

Mobile generators don't produce any more pollution than regular power plants. It is common for businesses to install mobile generators so that they can operate them right away (I believe the law allows them to operate for 1 year) before they get permanent air permits for them. By the way, Xai does have stationary permits for them now, so even that deeply flawed information is out of date.

It is hilarious to see morons like you vacillate between claiming these companies are irresponsible for using the grid without paying for new generation, and complaining when they do add the grid capacity to cover their use. It couldn't be more obvious that it has triggered a knee-jerk anti-development instinct in your lizard brain and that you have not capacity whatsoever to consider these developments rationally.

All this anti-datacenter nonsense is entirely unfounded in reality, so you are just another one of the stupid voters I am talking about.

It sounds like a good way to tank the economy forever, but ok.