RSS

abhayesian

Karma: 213

Trying to become a shoggoth whisperer

Align­ment Fak­ing Re­vis­ited: Im­proved Clas­sifiers and Open Source Extensions

Apr 8, 2025, 5:32 PM
144 points
19 comments12 min readLW link

Find­ing Back­ward Chain­ing Cir­cuits in Trans­form­ers Trained on Tree Search

May 28, 2024, 5:29 AM
50 points
1 comment9 min readLW link
(arxiv.org)