krona

It has zero network effects and barely any economies of scale. It's a good product without a moat which will get replaced by any company with their own compute and cheaper energy.

I think the only way out for Anthropic is to buy a few neoclouds as they go under, but even then it's a coin flip as to if they survive the next 3 years.

> LLMs are dumb and unpredictable

Dumb, yes. But if the LLM hacks a Korean bank so it can 'fix' your router misconfiguration, it's only unpredictable from your perspective, not the person who designed the training pipeline.

yeah i trained a 1000 param LLM in my browser literally yesterday. it's game over for Anthropic /s.

There are cross-cultural differences in moral reasoning, and I think this is very interesting. But that seems to make training a hyper-rational AI, with no meaningful theory of mind, to be 'moral' or 'just' even harder.

> What is today a moral imperative is not the same as what was moral 50/100 years ago and is not the same as what will be morally acceptable 50/100 years from now.

This is a fashionable 'blank-slatist' statement about moral reasoning which became popular a few decades ago.

> Logic would prefer moral derivations over feelings, stakes over etiquette, and a constant scheduled updating of morality based on the derivatives rather than holding onto the rot and pushing back against change as long as possible.

Humans haven't changed in any meaningful way in hundreds of thousands of years, and you're saying the moral landscape is constantly changing? How is this not contradictory?

or network effects, or intangible assets (e.g. a brand, patents) or even economies of scale which they own and can leverage.

Narrator: But this was a fiction. In reality, evidence of any profits being made was extremely hard to come by.

There are dozens of ongoing wrongful death, manslaughter and negligence claims going through the courts. I won't insult you by providing links.

> Nobody is talking about banning the sale of gas boilers.

Several energy suppliers websites have pages dedicated to this very topic.

The current 2035 target originates in the ministerial Heat and building strategy policy paper (2020):

"Our ambition is to phase out the installation of natural gas boilers beyond 2035."

In the Annex:

Aiming to phase out the installation of new and replacement natural gas boilers from 2035, once costs of low-carbon alternatives have fallen, in line with the natural replacement cycle, including any hydrogen-ready boilers in areas not converting to hydrogen, to ensure that all heating systems used in 2050 are compatible with Net Zero.

Given the energy policy of the UK, which to many engineers makes absolutely no sense at all, it's impossible to predict what will actually come out of the policy sausage factory next.

> "hey, timeshift your AC use and add some insulation to your house"

The UK has the oldest housing stock in Europe, and most likely in the world. [0]

Roughly 30% of the stock is single-skin, without a cavity. While it's not impossible to insulate such walls, it's generally inadvisable for many reasons. The same goes for the roofs of these houses; although you will find many companies willing to spray your roof with foam or other forms of thermal sealing, those roofs were never designed to be insulated, causing other potentially catastrophic problems once it's been done.

Generally, the poorest and oldest people in society occupy these houses, but no matter; the government will ban the sale of gas boilers to coerce them in to installing heat pumps, despite them being incompatible with the existing pipework and radiators.

0: https://files.bregroup.com/bretrust/The-Housing-Stock-of-the...

> “oh - use after free and out of bounds are trivial to solve” and ignore the enormously meaningful order of magnitude harder thread safety

This is what advocates of Rust are guilty of and CVE-2025-68260 is an example of it. Ultimately TSAN and KTSAN exist for a reason, and rust devs should learn how to use them effectively.

Compared to what? CVE-2025-68260?

> You can be annoyed by Rust fanboys, sure, but the more annoying are the Rust haters, because the hate always comes from a place of ignorance.

For what it's worth, Rust fanboys also seem just as vulnerable to their ignorance of C/C++ compiler extensions and tooling (e.g. sanitizers, bounds annotations, standard library hardening modes and so on.) which are just normal part of professional C/C++ use today.

"There are two ways of constructing a software design: One way is to make it so simple that there are obviously no deficiencies and the other way is to make it so complicated that there are no obvious deficiencies. The first method is far more difficult." - T. Hoare (1981)

Sure, a supermarket will sell avocados if they make a profit, and won't sell them if they don't. That's a very different business model; the product is the infrastructure, not the AI.

> Well there certainly is at least some kind of viable business running large AI models for a fee.

ok, where are:

- The economies of scale?

- The network effects?

- The switching costs?

- The intangible assets (e.g. brand?)

Running AI models for a fee has none of these. At best, there are some economies of scale for running a datacenter, but OpenAI and Anthropic have none.

The ancestry of the Victorian gentry can be traced back to the late medieval period (and some to the Norman conquest), many centuries before Britain had an empire.

The wealth of the merchant class also predated empire by centuries; the empire grew out of a series of alliances and partnerships of both ot these classes, via the landed gentry successfully co-opting mercantile wealth (e.g. see what happened to the East India Company).

Alphabet issued a very oversubscribed 100-year bond with 6.1% yield earlier this year to raise capital for datacenter expension.

Meanwhile, Anthropic/OpenAI will struggle to survive the next 24 months on their current trajectory.

> When you truly understand what you’re trying to accomplish, it’s easier to take a small step...

I think they're describing an approach to decision-making in the context of agents.

Engineers encountering LLMs for the first time think the problems they encounter are entirely novel, but really we are rehashing old lessons which Fred Brooks wrote about half a century ago, but in new jargon.

The essential difficulty of specifying what you actually want remains the same and nothing I've seen from LLMs will change that.

> we've built systems that are so goal-oriented, and so capable, that they will do almost anything...

I think you mean task oriented, because they're still generally terrible at goal oriented activities except in those domains where the goal can be reduced to a familiar, explicitly practiced task or pattern.

70% of Canadian exports are to the US. 6% to the EU. Why would e.g. Poland want to enter a trade war with the US on Canada's behalf?

Well it's an open question.

As is generally the case for dog owners whose dogs attack (sometimes kill) other people/animals. There would need to be a degree of negligence demonstrated (e.g. the dog was 'out of control' which has a specific legal criteria/threshold in the UK).

> Whether they have a soul or consciousness or feelings doesn't matter here

It does when it comes to accountability for what the model does. If the model is nothing more than the sum of its training data and regime, then the company (or individual) is responsible for its behaviour just like any other machine.

Few people think Waymo shouldn't have to take on the full liability risk of what it's cars do; it should be the same for LLMs.

Extreme temperature levels (>2.0) can push a GPT of its manifold, essentially producing predictions barely distinguishable from random noise (it flattens the probability distribution of the next token). In theory this could predict anything including the next field of mathematics (infinite monkey theorem) but realistically that would never happen.

However, how to we know the next field of mathematics isn't a novel combinations of several other sub-fields? That level of mathematics would be indistinguishable from magic to most people and so in their eyes the GPT did something truly inventive.

This is a self confession if I ever saw one.

Absolutely not. By most accounts they're terrible at fixing anything other than trivial bugs in complex codebases e.g. Linux kernel, but they're much better at finding them.

You're suggesting that LLMs get better at fixing bugs/vulnerabilities, but at the same time stop getting better at finding them? What if this difference is inherent and essential?

What contract? It's mostly corporate debt. Seems pretty plain and simple to me.

The contracts I guess you're talking about are compute obligations which frontier labs have. I don't need to know what those obligations are to know what happens to Oracle.

> The next problem is making the models efficient enough to run a profitable business.

In that world, what does Oracle do with a bunch of data centres which 1. It needs to pay for and 2. nobody needs. This is what the article is about: building supply far in advance of demand.

GPUs depreciate at a far, far faster rate than houses do. And then they need to be replaced. Who's paying for that? OpenAI and Anthropic don't even have positive free cash flow.