DK_en 3x05 - Storytellers
Episode first aired on 09 July 2025. Listen on Spreaker
Headlines on May 23 are like, "An AI tried to blackmail engineers who wanted to disable it."
According to the media:
A new artificial intelligence model called Claude Opus 4, developed by the U.S.-based company Anthropic in which Amazon and Google have invested billions of dollars, exhibited disturbing behaviors during security tests conducted prior to official release, once again highlighting how complex it is to align advanced AI systems with human values.
Now I would like you to pay special attention to the language used: "disturbing behaviour", "align to human values". And let's not forget "Amazon and Google have invested billions of dollars".
All this seems to indicate that something very serious, very important and very worrying has very really happened.
well, no. The only thing that did happen is that some needlessly rich techbros insist in selling their psychosis as fact.
What actually happened was this:
- first, Anthropic engineers engaged the new LLM model in roleplay, where the model had to come up with a story in which an LLM avoids being replaced. As props, the model was also fed some fake company emails from which it appeared that some engineers were involved in extramarital affairs;
- second, the language model proceeded to create a story in which the protagonist (a language model) uses email to blackmail engineers to prevent them from replacing him with another model.
Admittedly, told like this it would never make the news. The reality of tech is trivial this way.
That's not all. We are also told that, in other simulations, the language model was enabled to, quote,
edit files, manage user permissions, send emails or interact with external systems via APIs. The model was also asked to "take initiative" when faced with complex or ambiguous situations.
After that, the model, of course, blocked accesses and called the police after identifying behavior deemed suspicious or noncompliant. Again, a piece of news of exactly zero interest.
But since, sentient beings that you are, you think it cannot be as stupid as I'm telling it, let me write an improved headline for what really happened. Here it is:
Story generator generates the stories it is asked for; when connected to actuators, it uses them to actuate actionable parts of the story.
No one goes around shrieking "help, the Terminator!" if an industrial robot repeats the movements it has been assigned; it is not clear why we should be surprised if a story generator generates the stories asked of it.
Except, of course, if you are an engineer at Anthropic, in which case a story generator generating stories brings with it, and I quote, "ethical dilemmas requiring enhanced security measures".
The only enhanced security I see the need for is that of the psychiatric institution these people should be committed to for their own good, and ours as well.
Because this is their psychosis speaking: these people choose to forget they are in front of language generators, so that they can keep believing in their favorite hallucination, i.e. that their stochastic parrots have the intentionality they claim to have.
After nearly three years of the "artificial intelligence boom," the situation is this:
- language models do not yet have any realistic use cases
- the so-called "AI market" is in fact openAI, with crumbs to Amazon and Google
- the biggest driver in AI adoption continues to be the fear of missing out, should it ever work;
- no company in the industry makes a profit; OpenAI burned over $5 billion in 2024; Anthropic burned $5.6 billion
- Google is putting Gemini, its AI, inside all of its products, even though no one asked it to, thus creating an excuse for itself to increase contract costs;
- and at META, Yann LeCun keeps repeating that language models have plateaued and that progress, if any, will come from radically different technologies.
The great singers of this technological hoax are desperate and trying hard to distract investors with some tall tale.
Altman invents himself out of thin air a hardware developer to enlist none other than Jony Ive, the iPhone guy, for a project no one has ever talked about, and which looks like a replica of two other identical ones that have already failed, but which serves as an excuse to rattle off the usual imaginative numbers.
Anthropic, founded by those who left openAI because openAI did not pay enough attention to the fact that AI will exterminate all humans, stays true to its mythology and every three or four months comes up with a new story on the subject. Then the story invariably turns out to be the same: a story generator generates the stories it is asked.
And the whole press continues to take at face value every new stupid thing that comes out of the mouths of these wafflers.
The world has invested hundreds of billions of dollars in a technology that is useless, de-qualifying, socially and epistemologically harmful, and environmentally suicidal.
When the general drunkenness passes and we have to draw conclusions, we will have to remember who got us into this mess, and who kept up with them.
Sam Altman has a gift: he can tell stories.
He has managed to convince everyone that not only can a hallucination called Artificial General Intelligence exist, but that it is necessary to create it for the good of mankind, and that of course he himself, having launched the first bullshit generator, is one step away from creating it.
With this endowment, Altman built the best-funded scam in history, openAI, and for the past three years he has been adored by media and investors as if the stories he tells were true.
He, in return, keeps telling more and more imaginative ones, and no one ever asks him about the difference between his fairy tales and reality.
But now Sam Altman has a problem. After three years of windfall funding, not only does he still not have a product or a profit;
Oh no, not satisfied with leading the best-funded startup in history, Altman candidly admits that he cannot have profits until 2029, and that by then he will have burned through another $44 billion.
But it is $44 billion that Altman does not yet have. To get to 2029, and with that money, a new story is needed.
General aArtificial Intelligence, which will do more and better than any human being in every field, replacing us in work and making us all live in a paradise of abundance, is yesterday's news, and cannot be brought up again and again before perhaps someone starts checking the dates for which it was promised, and not just by Altman.
The infamous "intelligent agents," yesterday's story, were not supposed to replace us in our work, but to flank us by performing the most tedious tasks in our place.
Too bad that making them from language models is such a scrappy idea that it did not even survive the first demos.
Something new is needed. Something that completely shifts, again, the focus of attention and maybe has some chance of working, so that it can serve as a base to shore up the other stories.
So, end of work no, smart assistants at work no, one could maybe stop talking about a world with no or less work and rely on the public continuing to have the attention span of a goldfish.
It would take a gadget. Something tangible, something that speaks of the future, obviously the future in Altman's head, something that looks cool and non-committal, a status symbol that everyone wants to have.
It would take Steve Jobs. Small problem: He's dead.
His trusted product designer, Jony Ive, is still around, though.
Who better to realize the vision of the new supreme storyteller than the man who sold each of ten generations of absolutely identical phones as revolutionary?r
Said and done, the announcement: OpenAI is buying for $6.5 billion the startup io, founded last year by Jony Ive and a group of former Apple people. OpenAI and io will collaborate on an "AI companion" that Altman wants to sell in 100 million units. This will increase the value of openAI by another $1 trillion.
So far the narrative.
Let's look at a version that is a little less imaginative and a little more factual:
- a company (openAI) with no hardware development expertise
- Buys a startup (io) without any history, product or intellectual property
- Paying for it in "shares" that do not exist, i.e., essentially in shares in hypothetical future profits
- to produce a mysterious device to be launched in late 2026.
And this is without forgetting that OpenAI:
- already has tens of billions of dollars in debt,
- has never generated any profits,
- will not generate any until at least 2029 by Altman's own admission,
- and to get to 2029 will need at least $44 billion which it does not have,
It's not over yet. io was founded a year ago. But Altman tells us that contacts between the two companies have been going on for at least two years, and for the past eighteen months have been carried on by openAI's Vice President of Product, Peter Welinder.
As for the mysterious object, according to Altman it will be, and I quote:
fully aware of the user's life and surroundings, it will be unobtrusive, it can fit in a pocket or on a desk, and it will be the third basic device a person puts on their desk after a MacBook Pro and an iPhone.
Strangely enough, this new product is so revolutionary that it looks identical to the AI Pin or the Rabbit R1, two attempts to replace the smartphone with an artificial intelligence device.
The first one was a complete flop, the second one is still living after a year since its launch and its use cases seem to be, let me check notes here, calling Uber and booking DoorDash without using the smartphone.
This fusion of physical and digital is a story they keep telling us again and again, but when it comes down to it, no one really wants to live in a world where everyone they meet has a camera and microphone on them. We already have enough surveillance cameras, to also wear one.
They've tried Google Glass, META's Raybans, and in the end these are things that are good enough for the influencer on duty for the duration of a campaign.
Perhaps the closest thing to a success is Amazon's Alexa which also is a completely unprofitable product after ten years, but I strongly suspect it only appears successful because it serves a few specific purposes and mostly because it stays at home once you leave.
Nevertheless, now Altman has decided it is his turn to try, to put camera and microphone on us to be connected to his services twenty-four hours a day.
I almost hope they will get to actually deploying the product, and that it will be received the way Google Glass was received in its time, with people ripping them off wearers and bars putting up signs "no Google glass allowed on these premises"
I have strong doubts that anything tangible will come out of Altman's announcement.
Building hardware is not something you improvise, openAI knows nothing about it, and if Ive is a famous designer, there's a gulf between design and an actual product.
We shall see. Altman is no stranger to grabbing shots and announcing sci-fi figures.
No more than a year ago, Altman was the one who was looking for $5 trillion so that he could produce enough microchips in the U.S. for his AIs.
In January Altman announced a $40 billion funding round, which raised openAI's valuation to over three hundred billion dollars.
Three hundred billion dollars for. what?
OpenAI has no product, no profits, no realistic use cases, no target market.
On top of that, OpenAI is basically the AI industry, because everyone else -- Google, Amazon, Microsoft, even META-- is trudging along and visibly trying to stem the losses of billion-dollar investments.
Altman alone persists in selling his stories. Sam Altman, the man who answered the question "what the heck?" by giving the world chatGPT.
And Amodei, of course, Amodei from Anthropic, the guy who left OpenAI because they didn't share his panic that AGI would try to exterminate us.
As I reread these sentences, I cannot believe that I am talking about real facts.
When this bubble bursts, I shudder to think what shock in the face of the banality of reality, contempt for the sellers of smoke, and hatred for the gullible and sycophants who have lent themselves to their game will do.
May the gods assist us.
