Cam comments on Pretraining on Aligned AI Data Dramatically Reduces Misalignment—Even After Post-Training

[Cam]

Thanks for what I believe to be the most detailed and thoughtful overview of the alignment pretretraining field to date. I have been referring to this document while charting out our next steps, and I expect others interested in this direction should do the same!

[RogerDearnaley]

Thanks! I and others turned up a lot, and now I have a lot of reading to do: I thought rather than just making a reading list for myself, I might as well write it up as a post.