scasper(Stephen Casper)

Analo­gies be­tween scal­ing labs and mis­al­igned su­per­in­tel­li­gent AI

Deep For­get­ting & Un­learn­ing for Safely-Scoped LLMs

Scal­able And Trans­fer­able Black-Box Jailbreaks For Lan­guage Models Via Per­sona Modulation

The 6D effect: When com­pa­nies take risks, one email can be very pow­er­ful.

An­nounc­ing the CNN In­ter­pretabil­ity Competition

Open Prob­lems and Fun­da­men­tal Limi­ta­tions of RLHF

A Short Memo on AI In­ter­pretabil­ity Rain­bows

Ex­am­ples of Prompts that Make GPT-4 Out­put Falsehoods

Eight Strate­gies for Tack­ling the Hard Part of the Align­ment Problem

Take­aways from the Mechanis­tic In­ter­pretabil­ity Challenges

Ad­vice for En­ter­ing AI Safety Research

GPT-4 is eas­ily con­trol­led/​ex­ploited with tricky de­ci­sion the­o­retic dilem­mas.

EIS XII: Sum­mary

EIS XI: Mov­ing Forward

EIS X: Con­tinual Learn­ing, Mo­du­lar­ity, Com­pres­sion, and Biolog­i­cal Brains

EIS IX: In­ter­pretabil­ity and Adversaries

EIS VIII: An Eng­ineer’s Un­der­stand­ing of De­cep­tive Alignment

EIS VII: A Challenge for Mechanists

EIS VI: Cri­tiques of Mechanis­tic In­ter­pretabil­ity Work in AI Safety

EIS V: Blind Spots In AI Safety In­ter­pretabil­ity Research