jsteinhardt(Jacob Steinhardt)

What will GPT-2030 look like?

More Is Differ­ent for AI

AI Fore­cast­ing: One Year In

Fu­ture ML Sys­tems Will Be Qual­i­ta­tively Different

How much slower is re­mote work?

De­liber­ate Play

Early 2022 Paper Round-up

A Fer­vent Defense of Fre­quen­tist Statistics

AI Fore­cast­ing: Two Years In

Emer­gent De­cep­tion and Emer­gent Optimization

Prin­ci­ples for Pro­duc­tive Group Meetings

The Power of Noise

Un­solved ML Safety Problems

ML Sys­tems Will Have Weird Failure Modes

Beyond Bayesi­ans and Frequentists

How fast can we perform a for­ward pass?

Film Study for Research

GPT-2030 and Catas­trophic Drives: Four Vignettes

Mea­sure­ment, Op­ti­miza­tion, and Take-off Speed

Em­piri­cal Find­ings Gen­er­al­ize Sur­pris­ingly Far