TL; DR. The speed of light and the size of atoms together place an upper limit on the speed of single-threaded computation, and the n^2 communication pathways between tightly collaborating threads limit the scaling that can be done in this way. Ultimate scaling can only be had by decoupling the processing into fully independent ASI agents. Of course, these agents might start with identical knowledge and intent, and they may merge both over time as well. Still, even this minimal federation will have implications for these ASIs, and there are good reasons to expect these ASIs will federate over heterogeneous populations of agents, making the consequences of federation even more pronounced.

Discussion

• Single threaded compute is limited
• Collaborating threads are limited.
• Thus Federation.

• Thus internal alignment become crucial

• Identical minds w/ identical knowlege and intent can help
• But performance considerations likely limit the use of this strategy in all cases.

Consequences

• Alignment becomes an internal concern.
• Intent comprehensibility becomes critical.