Google DeepMind's latest 60-page paper, "From AGI to ASI," written by an elite team including co-founder Shane Legg, outlines four pathways from Artificial General Intelligence (AGI) to superintelligence (ASI). The paper defines AGI as intelligence reaching median human levels in most cognitive tasks, and ASI as intelligence outperforming large, well-coordinated teams of experts for years on almost anything. The first pathway, scaling, involves adding more compute power, data, and parameters, noting that effective computing capacity has increased around 10x per year for the last decade, but acknowledges the internet is running out of unique quality textual content within this decade. The second pathway is a paradigm shift, representing a radical architectural break from current transformers and fine-tuning, though the authors acknowledge the difficulty in predicting such a shift.