Basic Facts about Language Model Internals — AI Alignment Forum