I’m currently focused on AI risk, especially existential risk, using prompt injections and jailbreaks as a model for studying a wide class of model failures and investigating how this could help with understanding the bounding, generalization, and other behaviors of much larger models.

I’m always happy to talk to people about their work and to share whatever advice I might have (please use discretion in following the advice of strangers on the internet, especially those who seem prominent for totally unclear reasons); and more generally about (meta)science, AI, history, future, and other fun topics.

I do my best to read all of the emails I get (and sometimes I succeed), so always feel free to send me a note via – alexey@guzey.com – or check out my twitter here: @alexeyguzey.

Subscribe to get updates (archive):

Sep 20, Secular Armageddon

Sep 13, Omens of exceptional talent

Aug 23, 16 questions posed & answered by Leslie Berlin's book "Troublemakers: Silicon Valley's Coming of Age"

Jul 31, 10 notes from watching pulp fiction yet again

Jul 27, questions

Jul 10, Advice from Tyler Cowen

personal favorites

most notable

Personal

Fiction & Art

notable

Full archive sorted by date