Really interesting and informative—thanks for writing this!
On midterm safety: I think this is underexplored and not discussed enough here.
I have been thinking that generating more examples of how midterm safety risks could be catastrophic might be helpful (e.g. fusion power is apparently bottlenecked by computing power and algorithmic advances). Btw I thought this video by Stuart Armstrong on specification gaming was really good for skeptical specialists.
Also, I feel like more writing on midterm safety could be combined with Greg Lewis’ post on GCRs importance from a person affecting view because that significantly relaxes assumptions about the importance of AI safety that might seem alien to skeptics.
Really interesting and informative—thanks for writing this!
On midterm safety: I think this is underexplored and not discussed enough here.
I have been thinking that generating more examples of how midterm safety risks could be catastrophic might be helpful (e.g. fusion power is apparently bottlenecked by computing power and algorithmic advances). Btw I thought this video by Stuart Armstrong on specification gaming was really good for skeptical specialists.
Also, I feel like more writing on midterm safety could be combined with Greg Lewis’ post on GCRs importance from a person affecting view because that significantly relaxes assumptions about the importance of AI safety that might seem alien to skeptics.