Cleo Nardo

Ar­gu­ments for P

I can’t think of great in­ter­ven­tions for en­sur­ing third-party model ac­cess.

Model ac­cess for third-par­ties — it’s a big deal!

Third-par­ties should fo­cus on scru­ti­nis­ing sys­tem cards

What did “schem­ing” and “mech in­terp” mean pre-2023?

How might out­siders make things go well?

Which goals ac­tu­ally mo­ti­vate de­cep­tive al­ign­ment?

Let’s have more par­tial in­sid­ers.

Out­siders should fo­cus on specs/​con­sti­tu­tions (among other things)

How do in­ten­tional se­cret loy­alties differ from other schemer mo­ti­va­tions?

9 kinds of hard-to-ver­ify tasks

Au­tomat­ing philos­o­phy if Ti­mothy Willi­am­son is correct

Pos­i­tive-sum in­ter­ac­tions be­tween play­ers with lin­ear util­ity in resources

Sa­cred val­ues of fu­ture AIs

En­sur­ing Safety in Mixed Deployment

In­tro­spec­tive RSI vs Ex­tro­spec­tive RSI

Fo­cus­ing on Flour­ish­ing Even When Sur­vival is Un­likely (Part I)

North Sen­tine­lese Post-Singularity

Strat­egy-Steal­ing Ar­gu­ment Against AI Dealmaking

A Very Sim­ple Model of AI Dealmaking