Parallelization constraints could delay a technological singularity [Linkpost]

Today, I'll be linking both a Epoch blog post and a paper called The Bounded Parallelizability of R&D: Theory and Application to AI , about how standard models of AI R&D and standard models of the possibility of an intelligence explosion like Eth and Davidson 2025 neglect the possibility of bottlenecks because technological progress isn't perfectly parallelizable. I'm mostly going to do a linkpost because I believe this isn't common knowledge in the community, but I will sprinkle in some of my own takes on parts of the paper and the blog post. One important point to make is that the criticisms only apply to the technological singularity. An economic singularity where robots begin to be good enough such that they can replace people at doing physical tasks, enabling more resources and cheaper robots, thus enabling more robots ad infinitum until it reaches infinity in finite time is still possible ala our own history of superexponential growth (though in our reality, things would slow down at some point, and my hot take is that once we start to have our first dyson swarm complete, we can't grow exponentially anymore.) (Tom Davidson does have good criticisms of the model, but the correct criticisms are about using the model to predict timelines, rather than takeoff.) Without further ado, let's begin. Paralleization technology In a conventional model, technological progress depends on two forces: (1) ideas get harder to find as the technological frontier advances, but (2) society…