avturchin comments on Oversight of Unsafe Systems via Dynamic Safety Envelopes

[avturchin]

I have been writing a long draft about “active AI boxing” for a while, but this idea is new for me.