Variations of the Garrabrant-inductor 2016-09-23T19:11:25.000Z

The shutdown problem 2015-07-15T14:18:15.000Z

Comment by IAFF-User-83 (Imported-IAFF-User-83) on Variations of the Garrabrant-inductor · 2016-09-28T11:11:27.000Z · LW · GW

Running the traders on some r or $\varphi$ might also be useful if we want a logical inductor that try to decide a particular statement $\varphi$ rather than try to decide all statements at the same time. If you want to decide $\varphi$ you could first run the traders on $\varphi$ and later run the traders on ${\varphi }_i$ with a probability proportional to $P_n\left({\varphi }_i\right)$'s influence on $P_n\left(\varphi \right)$. But I think it would be difficult to prove formally that such an approach would improve the inductors speed at deciding $\varphi$, and I don't know if it would work in practice.

Comment by IAFF-User-83 (Imported-IAFF-User-83) on Another view of quantilizers: avoiding Goodhart's Law · 2016-02-06T14:28:51.000Z · LW · GW

Not sure how closely related it is (I have not read through it), but here is another paper trying to fight Goodhart's law: http://arxiv.org/abs/1506.06980