Thanks for collecting works/discussions in this area and offering your own takes. It’s great to see more interest on how to improve AI safety besides keeping human control, and I hope the recent trend continues.
You have several links to Will MacAskill talking about working in this area, but didn’t link to the specific comment/shortform, only his overall “quick takes” page.
The important takeaway is that future AI-powered humans might set themselves up for cooperation failure by learning too much too quickly. This would be particularly tragic if it resulted in acausal conflict.
There’s too little in this section for me to understand how you arrived at this conclusion/concern. It might benefit from a bit more content or references. (Other sections may also benefit from this, but I’m already more familiar with those topics and so may not have noticed.)
Thanks for collecting works/discussions in this area and offering your own takes. It’s great to see more interest on how to improve AI safety besides keeping human control, and I hope the recent trend continues.
You have several links to Will MacAskill talking about working in this area, but didn’t link to the specific comment/shortform, only his overall “quick takes” page.
There’s too little in this section for me to understand how you arrived at this conclusion/concern. It might benefit from a bit more content or references. (Other sections may also benefit from this, but I’m already more familiar with those topics and so may not have noticed.)