RSS

Stuart_Armstrong(Stuart Armstrong)

Karma: 15,492

Why un­rig­gable *al­most* im­plies uninfluenceable

Stuart_Armstrong9 Apr 2021 17:07 UTC
11 points
0 comments4 min readLW link

A pos­si­ble prefer­ence algorithm

Stuart_Armstrong8 Apr 2021 18:25 UTC
22 points
0 comments4 min readLW link

If you don’t de­sign for ex­trap­o­la­tion, you’ll ex­trap­o­late poorly—pos­si­bly fatally

Stuart_Armstrong8 Apr 2021 18:10 UTC
17 points
0 comments4 min readLW link

Which coun­ter­fac­tu­als should an AI fol­low?

Stuart_Armstrong7 Apr 2021 16:47 UTC
19 points
5 comments7 min readLW link

Toy model of prefer­ence, bias, and ex­tra information

Stuart_Armstrong24 Mar 2021 10:14 UTC
9 points
0 comments4 min readLW link

Prefer­ences and bi­ases, the in­for­ma­tion argument

Stuart_Armstrong23 Mar 2021 12:44 UTC
14 points
5 comments1 min readLW link

Why sig­moids are so hard to predict

Stuart_Armstrong18 Mar 2021 18:21 UTC
41 points
6 comments5 min readLW link

Con­nect­ing the good reg­u­la­tor the­o­rem with se­man­tics and sym­bol grounding

Stuart_Armstrong4 Mar 2021 14:35 UTC
11 points
0 comments2 min readLW link