Clouded Judgement: From Data Quantity to Data Quality

The article explores the evolution of AI scaling laws, moving from pre-training compute and internet data to inference-time compute and domain-specific expert knowledge (PhD-level data). It argues that frontier data collection in specialized domains like finance, law, and medicine represents the next frontier for AI model improvement, creating significant market opportunities for data collection and labeling companies. The piece includes detailed SaaS valuation multiples and operating metrics benchmarks for cloud software companies.