A model that can design a better battery has done half the job. Someone still has to build the factory, run the line and ship the product. For all of history that someone has been a person, and people are the scarcest input there is.
The second half of the singularity removes that bottleneck. Machines that can see, grasp, build and repair, directed by minds that plan better than we do, turn an idea into a medicine, a power plant or a home in years rather than generations.
Robots follow the same laws
- 327robotics papers in the meta-analysis by Sartor & Thompson
- Fasterrobot performance improves with scale notably faster than language tasks
The scaling laws that made language models predictable turn out to hold for robots too. A meta-analysis of 327 papers found that robot models follow the same kind of power laws, and that robotic task performance improves with scale notably faster than language performance does. The evidence is earlier here, but it points the same way.
The industrial explosion
Forethought calls what comes next an industrial explosion: robots building robots and the factories that make them, with industrial output that could double every few years or even every few months. Sam Altman writes that 2027 may see robots that can do tasks in the real world, and that robots able to build other robots “aren't that far off.”
There is room for it. Humanity today produces the equivalent of roughly 0.01% of the energy in the sunlight that reaches Earth. The ceiling is not physics. It is how fast we can build.
“AI that can accelerate research could drive a century of technological progress over just a few years.”
Where robots are today
In Forethought's own words, general-purpose robots are “about as dextrous in general-purpose contexts as an extremely drunk person, but they are rapidly improving.”
We quote the whole sentence because the second half matters as much as the first. Language models were a curiosity five years before they were writing code.
We would rather be early to the robotic half of this than explain later why we missed it.