Calling out a few sage quotes from JT here: > Alignment is not as hard to solve as many claim, but it is in the end an algorithmic problem > Pretraining takes gradient with respect to next token prediction - that's def not the alignment objective. > In the end for humans we write a lengthy set of rules we want them to obey and we punish those that do not. When we’re unsure about what rules apply to that situation we have lengthy arguments about it and try to make a fair and just decision at…