The latest edition of Jack Clark's Import AI newsletter leads with three substantial items: an AI called Fable that writes GPU kernels, a broad look at AI-driven automation, and analog computation research. The headline asks whether this marks the beginning of a new world, framing the clusters as early signs of a deeper shift in how compute is produced and applied.
Fable's ability to generate GPU kernels stands out because writing such low-level code typically demands deep hardware expertise. If AI can reliably produce optimized kernels, it could ease a bottleneck that limits many machine learning and scientific workloads. The newsletter presents this as an example of AI moving into increasingly specialized engineering territory.
On automation, Import AI examines how AI is being used not just for isolated coding tasks but across larger workflows, potentially reshaping roles and productivity norms. The analog computation item explores hardware designs that process information in ways closer to physical signals, a departure from the dominant digital paradigm, which could offer efficiency gains in certain applications.
All three topics are drawn from the single source, so there is no competing viewpoint to compare. Rather, they collectively suggest that the next phase of AI may be defined as much by who builds the machinery as by the models themselves.