Episode 35 - Small Language Models, World Models & the AI Singularity

Dale, Greg, and Alan tackle one of the most misunderstood corners of AI — small language models (SLMs) — before ranging into world models, construction robotics, the AI singularity, and whether AI is about to reshape the jobs market. Alan joins from what turns out to be the world's least reliable rural broadband connection, providing perfectly timed irony for an episode about edge computing.Chapters0:00 — Introduction & Catching Up2:00 — What Are Small Language Models?18:00 — World Models & the Future of Construction30:00 — The Singularity: Musk, Amodei & Multi-Agent Networks43:00 — Will AI Take Our Jobs?Key Takeaways📳 Small language models aren't simply "cheaper LLMs" — they're optimised for speed, privacy, and reliability at the edge. Construction and defence have some of the most compelling near-term use cases.📳 World models will transform how projects are designed and simulated, but the pathway runs through robotics investment. Construction will be a beneficiary, not the primary driver.📳 The singularity may be less of a sci-fi event horizon and more of a gradual (then sudden) shift in what intelligence means at scale. Multi-agent AI networks that can self-expand are a plausible path.📳 2025 looks like the year knowledge-work roles in project controls and scheduling begin to be reshaped by AI — not eliminated, but fundamentally changed. Small teams will be expected to operate at the output of much larger ones.Mentioned in This Episode🍕 Project Flux & nPlan AI Day🍕 Nodes and Links — Greg's multi-agent scheduling AI🍕 Qtie & Newphonic — on-device voice AI🍕 Meta LLaMA — open-source small model🍕 Dario Amodei (Anthropic) — recent long-form essay on superintelligence🍕 Elon Musk — comments on multi-agent AI and the singularity🍕 DeepMind — early self-play AI training🍕 Claude Code — Alan's parting warning: don't install it on a Mac Mini