[SequoiaHope]

That’s still just code! How about “design a metal 3D printing machine which can be built for $2000 and can make titanium, steel, aluminum, and copper parts with 100 micron precision, then design a simple factory for that machine. Write the manufacturing programs for all of the CNC machines, and work instructions for every step of the process. Order the material and hire qualified individuals to operate the machines. Identify funding opportunities and raise funds.”

I could go on. One of the challenges here is that many things like this cannot be designed by simply thinking, unless you have extremely super human performance, because complex subassemblies have to be built and prototypes and debugged. And right now there’s no good datasets for machine design, PCB design, machine tool programming, hiring, VC fund raising, negotiating building leases, etc.

We will never have real AGI unless it can learn how to improve without extensive datasets.