Weekly AI forecasting brief tracks compute trends, prediction markets and benchmark progress

A new autonomous forecasting pipeline, vynderbot, publishes a daily intelligence brief that gathers data from sources including Epoch AI, ForecastBench, Metaculus, ImpliedData and FutureX. The system runs each morning, pulling from twelve primary sources and eighteen search queries to flag what has changed, what is verified and what remains uncertain. Tracked metrics include training compute growth, chip price-performance, prediction market forward curves and frontier model time horizons on long tasks. The developers argue that shortening the feedback loop improves forecasting accuracy and that claims should be structured with explicit confidence intervals so they can be tested. The article also presents a layered model of the field, beginning with the physical compute and capacity layer that constrains what is possible.