Dmitry Vaintrob

Steel­man­ning heuris­tic arguments

Me­moriza­tion-gen­er­al­iza­tion in practice

Effi­ciency spec­tra and “bucket of cir­cuits” cartoons

The mem­o­riza­tion-gen­er­al­iza­tion spec­trum and learn­ing coefficients

My su­pervillain ori­gin story

The gen­er­al­iza­tion phase diagram

On polytopes

QFT and neu­ral nets: the ba­sic idea

Writ­ing ex­per­i­ments and the ba­nana es­cape valve

Against blan­ket ar­gu­ments against interpretability

Log­its, log-odds, and loss for par­allel circuits

Is the­ory good or bad for AI safety?

Renor­mal­iza­tion Re­dux: QFT Tech­niques for AI Interpretability

The quan­tum red pill or: They lied to you, we live in the (den­sity) matrix

Per­ma­nents: much more than you wanted to know

Ba­sics of Bayesian learning

Paper club: He et al. on mod­u­lar ar­ith­metic (part I)

The pur­pose­ful drunkard

Dmitry’s Koan

The ab­solute ba­sics of rep­re­sen­ta­tion the­ory of finite groups