Shortcut to full PDF: https://arxiv.org/pdf/1705.03394.pdf https://arxiv.org/pdf/1705.03394.pdf On a first reading, and accepting the major premises for the s

Shortcut to full PDF: https://arxiv.org/pdf/1705.03394.pdf https://arxiv.org/pdf/1705.03394.pdf On a first reading, and accepting the major premises for the sake of argument, this seemed like the weakest point: The reader might wonder whether starting these computations now is rational since the universe is quickly cooling and will soon (compared to the overall lifespan of the civilization) reach convenient temperatures. The computational gain of doing computations at time t is ∝ exp(Ht): it increases exponentially until the temperature is dominated by the internal heating rather than the outside temperature. Since most of the integrated value accrues within the last e-folding and the energy used early was used exponentially inefficiently, it is not worth starting early even if the wait is a minuscule fraction of the overall lifespan. Grant that there's an exponential efficiency gap between computing now and computing a trillion+ years from now. There is also an exponential efficiency gap between using the power radiated from stars to power Dyson Swarm computing for the next trillion years and just letting it radiate away without computing anything, like a chump. A coolly rational maximizer isn't going to leave additional gains on the table with the human-like reasoning of "it's only a very small additional gain."