A speculative chipmaking path to rapid AI self-improvement

A poster argues that high-throughput, easily scalable atomic or molecular assembly for custom “chips” could make a rapid AI takeoff possible. As an illustration, they point to Richard Ngo’s “Tinker,” a hard-sci-fi sketch of a modestly superintelligent agent reaching a qualitatively better computing substrate. They say the sketch is probably wrong, though something similar might work.