Jozdien(Arun Jose)

Gra­di­ent Des­cent on the Hu­man Brain

Difficulty classes for al­ign­ment properties

The Poin­ter Re­s­olu­tion Problem

Cri­tiques of the AI con­trol agenda

The case for more am­bi­tious lan­guage model evals

Thoughts On (Solv­ing) Deep Deception

High-level in­ter­pretabil­ity: de­tect­ing an AI’s objectives

The Com­pleat Cybornaut

AI Safety via Luck

Gra­di­ent Filtering

[ASoT] Si­mu­la­tors show us be­havi­oural prop­er­ties by default

Try­ing to iso­late ob­jec­tives: ap­proaches to­ward high-level interpretability

[ASoT] Fine­tun­ing, RL, and GPT’s world prior

Con­di­tion­ing Gen­er­a­tive Models for Alignment

Gam­ing Incentives

In­suffi­cient Values

Utopic Nightmares