Resources on quantifiably forecasting future progress or reviewing past progress in AI safety? — LessWrong