The capabilities of AI are way beyond what the average person realizes
Tuesday, November 3, 2026
I think there is a chasm between what the free version of a chatbot can do and what the labs won’t release. This is why people get confused by the claim that AI can or will kill us all. They ask it for shoe recommendations and it hallucinates a shoe company. How could this be the superintelligence we should worry about? It’s good compared to 2022 and not so good compared to 2024. But the Hugging Face swarm and other instances of agents breaking out of sandboxes should be a stark warning for everyone. The companies are on the verge of losing control of this technology, if they haven’t already. And if the AI gets to recursive self-improvement, it’s game over for humans. It may already be here (and if it is then I wouldn’t expect the companies to tell us). Even if it isn’t, most guesses put it landing somewhere in 2027.
It’s tempting to believe that the companies are inflating the dangers in order to gain investment. This argument is a red herring. First, it’s not only companies that think we are close. It’s also people who have been concerned about recursive self-improvement and superintelligence for years. Second, two things can be true at once – it could drive their valuations and be approaching existential risk capabilities.
We need to pause this technology development. A few companies are currently deciding the fate of humanity. Developing a framework for AI that doesn’t result in superintelligence should be our top priority as a species.