Conditioning Predictive Models: Open problems, Conclusion, and Appendix — LessWrong