The AI Forecasting Agenda
The AI Forecasting Agenda
AI forecasters now match or beat top human performance on traditional forecasting questions, and foundation models like ChatGPT and Claude are roughly six months behind specialized systems, so they may catch up by year's end. The piece surveys open challenges from the Forecasting Research Institute, including conditional and causal forecasting, forecasting AI progress, long-term forecasts, unresolvable epistemic questions, question generation, and whether AIs can break from the respectable liberal consensus. The author argues for using AI forecasters now, citing a FutureSearch estimate that Anthropic disbanding would raise the chance of a global pause from 16% to 19%, while acknowledging this rests on unproven conditional, causal, and AI-progress forecasting.
Source: Hacker News AI · astralcodexten.com