Stuart_Armstrong comments on Value extrapolation partially resolves symbol grounding

[Stuart_Armstrong]

Thanks—this feels somewhat similar to my vague idea of “define wireheading, tell AI to avoid situations like that”.