According to Andrej Karpathy (@karpathy), a new project used the GPT-5.1 Thinking API to conduct an in-hindsight analysis of 930 frontpage Hacker News articles and discussions from December 2015, automatically grading comments based on their predictive accuracy with today's knowledge (source: @karpathy, karpathy.bearblog.dev/auto-grade-hn/). The process took approximately 3 hours of coding, 1 hour to run, and cost $60 in API usage, demonstrating the efficiency and scalability of advanced LLMs for evaluating historical digital content. This approach highlights a practical application of AI in benchmarking foresight, training forward-prediction models, and extracting actionable insights from historical data. The project showcases significant business opportunities for AI in content analysis, reputation scoring, and automated knowledge mining, pointing to a future where LLMs can cheaply and accurately scrutinize vast internet archives for strategic and commercial value (source: @karpathy, github.com/karpathy/hn-time-capsule).