PraCha, home

AI Safety, Alignment, Human & AI Systems, Evals, Diffusion and Futures

Build and evaluate capable AI systems that are safe, aligned, human-centered and understood within broader technological and societal futures.

courseplannedphilosophy · science

An engraved elevation of an ornate wooden lighthouse on a rock in the sea, with galleries, a lantern and cranes.
South elevation of Winstanley's first Eddystone lighthouse, plate 4 of A Narrative of the Building of the Edystone Lighthouse, John Smeaton, after a drawing by Jaaziell Johnston, 1813 edition (first published 1791). State Library of New South Wales, RB/F656.599/1. Public domain.
For
Anyone building, deploying or governing advanced AI
You will be able to
  • Explain the core alignment problems and the main approaches to them
  • Evaluate capabilities and risks with measurements that mean what they claim
  • Use interpretability tools to test claims about what a model is doing
  • Analyse how humans and AI systems shape each other, and where oversight fails
  • Trace how AI diffuses through institutions and societies, and its second-order effects
  • Reason about long-horizon risk — collapse and catastrophe — and what reduces it

Questions about this course, or a case that argues with it? Write to mepc3197@columbia.edu · the title goes in the subject line

Related

Sent through Substack, which will ask you to confirm. Your email is used for these letters and nothing else. Subscribe on Substack instead.