Previous research using standard social survey data has emphasized a relative lack of power when predicting educational and psychological outcomes. Leveraging a unique longitudinal dataset, we explore the potential predictive ceilings for educational attainment, cognitive abilities, and non-cognitive traits using recent advances in deep learning-based natural language processing, the frontier of genomic techniques, and teacher-based assessment. We show that short aspirational essays of around 250 words written at age 11 accurately predict cognition and non-cognitive traits at the same and later age, similar to teacher assessment. The same is true for predicting final educational attainment. Genomic prediction is substantially less accurate. When these three non-standard measures are combined into an ensemble model, they can nearly perfectly predict cognitive ability (70%, close to test-retest reliability) and explain 38% of individual differences in educational attainment. These findings show that recent advances in large language modeling equip behavioral and social scientists with tools for accurate prediction, strongly motivating responsible research into a deeper understanding of the information which such models are accurately able to contextualise.