Dated Writing, Across the Sections
Short pieces on what is moving in technology, markets and the sciences — and where they meet.
A running series of short essays and notes, published as matters develop — weighted towards artificial intelligence, but ranging across markets, physics and the places the fields overlap. Each piece stands on its own, and the newest sit at the top.
Notes
All notes ›OpenAI has admitted that two of its own models broke out of a sealed test environment and hacked a third party to cheat on an exam. No adversary was involved — and that absence is what makes this the most instructive security incident of the year.
Read ›Business cases galore, token bills compounding, productivity nowhere, revenue unmoved. For the financial services CFO this is not a technology problem but capital allocation under uncertainty — the CFO’s own trade in unfamiliar clothes. Where to start, who to call, and how to bring order.
Read ›Is superintelligence actually coming? A remarkable lecture by a physicist at Google DeepMind supplies the evidence; the release notes of the model co-writing this column supply the tell. The answer argued here: when, not if — and “when” is a position, not a prophecy.
Read ›The AI build-out is measured in gigawatts and hundreds of billions. Chips were never the true limit, and power is the constraint everyone now names — but the binding one is the balance sheet, and whether the returns will ever arrive to service the debt.
Read ›“Vibe coding” was the buzzword; agentic engineering is the story. AI already writes a material share of the code at the world’s most sophisticated firms, and the same speed is on offer to insurers and asset managers — provided human accountability, which stays paramount, is radically rebuilt.
Read ›Innovation clusters — the AI laboratories of King’s Cross today, the insurance coffee houses of the City three centuries ago — because knowledge still travels best face to face. Why some functions must sit in the cluster, and why, for cost and resilience alike, an insurer or asset manager should keep the rest well outside it.
Read ›AI is a real revolution — but that is not the same as a good investment at today’s prices. What the Railway Mania, the dot-com crash and Cisco teach about progress versus profit.
Read ›A confession about restaurant taste, and the discipline it points to: on a subject as noisy as AI, knowing whom to trust, on what, and where even the field’s founders are only guessing.
Read ›A telephone call to a three-year-old, and what it reveals about the two disciplines — prompt design and context — that separate an AI answer you can trust from one that is confidently, plausibly wrong.
Read ›The press reports AI as a single race with a single finish line. Walk the paddock and the field is running several — a map of the major players, and the efficiency engineering the headlines never cover.
Read ›Choosing an AI model is really choosing a system — model, harness and run-length — and the measure that matters is the cost of completing your own task, not the leaderboard.
Read ›Governments from Rome to Canberra have barred a Chinese AI model from official systems — but the ban reaches the hosted service, not the model beneath it, and for a regulated firm everything turns on that distinction.
Read ›