MIRI announces a new \”Death With Dignity Strategy\”
Tl;dr. It’s clear that at this point, humanity will not solve the alignment issue, nor will it even try to do so, or fight for its survival. We should focus our efforts on helping humanity to die with a little more dignity, since survival is not possible.
Let’s not be shy. MIRI did not solve AGI alignment, and it knows this. Paul Christiano’s complex schemes will never work in the real world before DeepMind destroys it. Chris Olah’s transparency work will, at the current rate of progress, allow someone at DeepMind to give a highly speculation warning about how a set of huge inscrutable Tensors inside a system, which was recompiled 3 weeks ago, and has been training by gradient descend for 20 days, may possibly plan to try to deceive their operators.
The management will ask them what they should do.