Conditioning Predictive Models: Large language models as predictors — AI Alignment Forum