Work on Bayesian fitting of AI trends of performance? — LessWrong