János Kramár(János Kramár)

AtP*: An effi­cient and scal­able method for lo­cal­iz­ing LLM be­havi­our to components

Fact Find­ing: Do Early Lay­ers Spe­cial­ise in Lo­cal Pro­cess­ing? (Post 5)

Fact Find­ing: How to Think About In­ter­pret­ing Me­mori­sa­tion (Post 4)

Fact Find­ing: Try­ing to Mechanis­ti­cally Un­der­stand­ing Early MLPs (Post 3)

Fact Find­ing: Sim­plify­ing the Cir­cuit (Post 2)

Fact Find­ing: At­tempt­ing to Re­v­erse-Eng­ineer Fac­tual Re­call on the Neu­ron Level (Post 1)

Does Cir­cuit Anal­y­sis In­ter­pretabil­ity Scale? Ev­i­dence from Mul­ti­ple Choice Ca­pa­bil­ities in Chinchilla

In­finite Mo­dal Com­bat: some observations

A tractable, in­ter­pretable for­mu­la­tion of ap­prox­i­mate con­di­tion­ing for pair­wise-speci­fied prob­a­bil­ity dis­tri­bu­tions over truth values