They are pulling your leg. When CEOs, and political figures too, say something in public, one mistake you absolutely must not make is to take what they say literally and then parrot it back, especially if there is a following on social media involved, or if you want to look good in front of your friends. It is the opposite. When these people say something, you have to understand the subtext.
I will take a few of Anthropic’s statements from recent months as an example, to work out the underlying arc, because in my opinion a lot of people just do not see it - partly because they are distracted, going “look at this stuff, in six months developers won’t have jobs anymore.”
Strategy one: take the opposite of what they say
Take the opposite of what they say and you get: in six months developers will still be working. The truth is often the opposite of what they say, because why did they say it? These are actors. They have a program, an agenda. They have people working for them who write what they should say. So the opposite of what they say often leads directly to the truth, because the fact that they are saying it is the justification of something. If someone comes out and says X, they are often trying to push X, which means X is not true. That is the first strategy.
Amodei says: in six months developers won’t have jobs. I negate it - developers will be working, and how. But he says it in the context of his own interest.
Strategy two: what interest do they have in saying it?
What interest does this person have in saying that thing? His interest is to sell his artificial intelligence services, so he plays on emotion, plays on fear, and above all plays on an implicit comparison between what his services do and what people do. He does not give a damn about people. What he is really saying is a message aimed above all at companies, telling them: “Watch out, I cost you a lot less, and what I offer as an AI service lets you pay fewer people, maybe more specialized, but fewer people on the payroll. So in six months you’ll be able to have smaller teams and keep productivity unchanged.” That is the message behind “in six months developers won’t have jobs,” and it is all the more true if you analyze the interests of the person making the statement. If you look at Anthropic’s history and commercial strategy, you will notice it has specialized - especially in competition with OpenAI - in serving companies.
Strategy three: the historical context
Three: what is the historical context? Why does this statement arrive now? They say things when they need to say them, because it serves them. And anyone who repeats or literally interprets what they say, even relaying it as if it were big news - including the whole class of tech journalists, who I consider largely incompetent because they never make interpretations of this kind and just broadcast “ok, this thing happened,” pure advertising - is missing the point.
The Mythos example
The other example is Mythos. In this phase I am talking about Anthropic because Anthropic is doing well, but OpenAI, in the three or four years before Anthropic’s overtaking, gave us dozens of these. Sam Altman is the Silicon Valley equivalent of the satanic version of Pinocchio: when Altman opens his mouth, it is surely the opposite of what he says. In that, he is extremely reliable. So I do not have it in for Anthropic - in my opinion they are the least bad, and their products are excellent. I am talking about commercial propaganda, and I repeat, all of this applies to politics too.
When you hear “we built the most powerful model of all time, it is called Mythos, here is a report of hundreds of pages detailing what it does, but we are not releasing it because it is too dangerous” - sigh. So, the opposite of what they are saying is: it is not dangerous and we are releasing it. Which is true, because what did they do? They took this model, which fine, may well be the most powerful of all, and they released it - but only to Enterprise, only to the biggest companies. They launched a program with this halo, this cover of security and safety, and they are starting to let only Enterprise use Mythos.
So, negation: it is not true that it is so much more dangerous than the hundreds of models already running. It is not true that they are not releasing it. And interpreting the exact opposite of what they said has already brought us closer to the truth.
Now, what interest do they have in saying “it is too dangerous, we are not releasing it”? What they actually mean is: we are releasing it to a group of customers who are worth more to us. What they are saying is: this is a really cool thing, but we are not giving it to everyone, we are giving it only where the money is, which is Enterprise, where B2B pays much more. Anthropic’s general strategy in recent years has been to specialize in enterprises, professionals and companies - which, among other things, was a smart way to compete with OpenAI, which instead scattered itself across dozens of consumer offerings. Consumers tend to be much more expensive, especially to acquire: it takes investment to acquire customers and they pay little anyway. Companies, on the other hand, not only pay more, there is also friction - once you have contractualized things it takes a lot of time to switch, so once you are in, you stay. So the truth of Mythos is that it is the first high-end AI for companies, for large enterprises.
Why don’t they give it to everyone?
Another question along these lines: why don’t they give it to everyone anyway? They give it to enterprises, charge more, and then maybe let others use it too, right? Here too, what do these CEOs never say, but which is the subtext across the whole sector? The essential reason they wage war on each other, bicker, and strike circular trillion-dollar deals is that this stuff costs - at the infrastructure level, the data center level, the energy level, the data level.
The reason Amodei founded Anthropic by breaking away from OpenAI - because they were all at OpenAI first, and Sutskever broke away too - is an internal fight they had from the earliest days about compute: do we dedicate more compute to the product, to sell, or do we dedicate more compute to research and development? Since Amodei in there was tied to R&D and the safety department, it clashed with Altman, who is far more commercial. They fought over who should have more infrastructure available, and at a certain point Amodei left. Now that Amodei has investors breathing down his neck, he has to do the same things Altman did.
To recap: saying “we published Mythos, it is too dangerous, by the way here is the Enterprise program, we are not releasing it, but here is the Enterprise program” - the correct translation is: we do not have the infrastructure to run a beast like this for huge masses of people, so we limit our efforts to serving whoever pays the most. It is easy, and it is the exact opposite of what they said.
