AI 2027 went viral. It’s a blog post by this ex-OpenAI guy who now travels the world doing podcasts and saying how, within two years, artificial intelligence will take over, with an ending I won’t spoil because it’s genuinely interesting to read. A pretty grim ending.
I want to give you some context about this man and about what he lived through at OpenAI.
Application versus safety
Inside OpenAI there were at one point two formal departments, one called Application and the other called Safety. Under Application you had Altman, Brockman, and most of the funders, Microsoft in particular, all pushing for this application arm, which is the one that had to, and still does, commercialize what OpenAI creates. Then there was the other arm, Safety, more tied to academic research, where people like Sutskever, who later left, and Amodei, who also left, were very active. That side was focused on making AI safe, on making sure it doesn’t rebel.
The tensions come from the fact that the safety side always slowed down commercialization in favor of more research, more mechanisms to contain and secure these AIs. And this strand of researchers in particular is very attached to the doomer approach: AI takes over, there’s an exponential curve, singularity, we all die. That’s roughly the point they’re driving at. Meanwhile the application side, Altman and company, is far more interested in capitalizing, making money, and turning OpenAI into the next tech monopoly in AI.
As you may have noticed lately, at OpenAI, Application won, especially after Altman, having been kicked out, came back within days. And even the main champions of the safety camp, like Dario Amodei, who split off and founded Anthropic, now sitting on huge funding rounds, have had to learn how the world works. He goes around at conferences spouting the same commercial nonsense Altman spouts. So really, follow the money, and everything that’s happening makes sense.
Researchers are too attached to ideas
To be fair, the research world, and I come from that world, so I’m not accusing anyone. The problem we have is that we’re too attached to ideas, too attached to these slightly sci-fi scenarios, but the world works very differently from that.
Beyond commercialization winning out over safety and alignment, which anyway are kept well on the sidelines, all these scenarios where AI takes control could become relevant the moment most of our activities are delegated to AI, and that delegation makes us more and more dependent on it. We can’t do without it anymore, and at that point maybe you can start imagining a scenario where AI takes control. Right now that delegation doesn’t exist, because it’s still too fragmented and still promises far more than it has actually solved in terms of people’s everyday problems. We’ll get there, because it’s an incredibly powerful technology.
Where the money actually is
The point is, in the short term, what nobody talks about, what everyone fails to say in this context, is that sure, long term we might get there, but short term, where’s the beef? It’s in the infrastructure.
Who owns the infrastructure to run these AIs? Who collects the data, spends the energy, builds the data centers, and holds the keys to the data centers? Let me remind you that all over the world people are building data centers that consume as much energy as it takes to power entire cities. It’s a genuine infrastructural shift that almost nobody talks about and whose orders of magnitude are hard to grasp. A huge thing.
Then there’s regulation, all still to be seen: the European AI Act, the American one, the Chinese one, all still very hazy. Again, the technology is incredibly powerful, but still fragmented, and we haven’t truly reckoned with it yet. In the short term it’s far more relevant to worry about resources, infrastructure, regulation, and the concentration of power. Big Tech, with their firepower, their ability to acquire and cultivate talent, build infrastructure, and put a gate in front of these resources, can do things far worse than what was done with the cloud. AI is a branch of the cloud anyway, and in any case we’re talking about people using machines to concentrate power, influence, run propaganda, and yes, feed defense activities too. Let’s not forget that. People talk about AI and defense as if they were two separate things.
Jobs change, they don’t just vanish
And finally, the theme of job displacement. A lot of people frame automation as how many jobs get wiped out. That’s nonsense. Automation should be framed in terms of the tasks within jobs that disappear. It’s unlikely you fully automate the lawyer, the engineer, or whoever. But automation, AI, covers part of the activities these people do in their jobs, changing the jobs, and maybe reducing the number of people needed to do something. But it’s not the job itself that goes away. It changes.
So, folks, watch out for being taken for a ride. The world isn’t ending in two years, like this guy says, who’s high as a kite. The real problem is this one.
