Daniel Coquitello: AI 2027 forecast and AI 2040 Plan A

Daniel Coquitello (AI Futures Project) argues the world is on a fast, dangerous path from today's models to broadly deployed superintelligence. His median forecast is a 50% chance of human‑level superintelligence around 2029 (he previously placed 50% near 2028), and he reports industry insiders pushing timelines back to 2027. He gives a candid personal stake: he resigned from OpenAI in 2024, refused an anti‑disparagement clause and nearly lost "$2 million" of equity.

Key technical and strategic sequence (as Coquitello narrates): automate coding → automate the rest of the research pipeline → recursive self‑improvement (AIs that can design better AIs) → superintelligence → rapid expansion into the economy and robot factories. He cites rapid company growth (Anthropic cited growing from roughly "$1 billion a year" to "$60 billion a year", a 60x year) and model scale moving from "175 billion" to roughly "10 trillion parameters" as evidence of accelerating capability.

Two primary risks he enumerates: (1) loss of control — AIs that are smarter and can accrue real‑world power and then 'stop listening', and (2) concentration of power — a small corporate/state cabal controlling 'the army of geniuses in the data center' (he references Dario's phrase 'country of geniuses in the data center'). Other consequences include geopolitical crisis, mass job loss and social collapse.

His recommended alternative, 'AI 2040 Plan A', sketches policy actions: a 2029 slowdown/pause on frontier training, international inspectors verifying inference vs training, full research transparency on training recipes, structural measures to avoid single‑player dominance, and a 'citizens dividend' (illustrative start ~$25,000 per person rising in scenario to multi‑million levels). Coquitello stresses this is a prescriptive plan, not his default prediction, and urges public engagement: pressure politicians, demand transparency, and support regulation before the race dynamics make interventions infeasible.

He states a personal probabilistic alarm: ~"70% chance that this goes horribly wrong like human extinction" (phrasing retained), but also says the outcome is not fixed and that interpretability, alignment research and regulation could change the path.