Follow your favourite podcasts, listen offline and in the car with CarPlay and Android Auto, and always pick up where you left off. Free to try.
Matt McDonagh is back for episode four of Life in the Singularity, and this week he pulls back the curtain on the difference between a model and the harness built around it. He explains why the model is probabilistic while the harness is deterministic, how guardrails keep outputs in bounds, and when it actually makes sense to fine-tune a model instead of leaning on the big general ones.
From there Matt opens up his own setup on screen: projects in Codex backed by GitHub, a research operation he calls the Compute Intelligence Agency that tracks the frontier labs, and a general manager thread on a heartbeat that directs research, commercial and red team agents. He makes the case for skill files as the newest step in humanity's long story of storing and sharing knowledge, from tribal elders to radio to the internet.
The back half plays a short clip of Demis Hassabis defining AGI and asks whether we are already past it, then shows off his persistent assistant Ember, NotebookLM's video tools, metamaterials, data center power and cooling, and why whoever learns fastest wins the singularity.
