Organisation · nonprofit · US

Alignment Research Center

A nonprofit doing theoretical AI alignment research — eliciting latent knowledge and heuristic arguments — whose evaluations team spun out to become METR.

1 entry · 25 May 2023

The Alignment Research Center (ARC) is a nonprofit founded in 2021 by Paul Christiano, a former OpenAI researcher, to work on the theoretical side of AI alignment — how to make advanced systems reliably do what their operators intend. Its best-known technical programme concerns "eliciting latent knowledge" and heuristic arguments for why a model behaves as it does. ARC's evaluations team, which ran early tests of frontier models for dangerous capabilities, spun out in 2023 to become the independent nonprofit METR, leaving ARC focused on conceptual research. It is a small Berkeley-based group whose influence rests on framing problems the rest of the field later takes up.

Category
Safety & alignment research
Founded
2021
HQ
Berkeley, US
Key people
Paul Christiano

Appears alongside

Featured in threads

Tracks

  • Safety & alignment 1

  1. DeepMind and collaborators publish framework for evaluating extreme AI risks

    Twenty-one researchers across nine labs and universities proposed testing models for capabilities such as deception and cyber-offence before training runs finish, not after release.

    Safety & alignment