What is an embedded evaluator, the top job AI chiefs are hiring for?
A hot job title, "embedded evaluator," has entered the AI space — and AI chiefs say these people will stand between humanity and the apocalypse.
TL;DR
- Anthropic CEO Dario Amodei suggests embedding independent "embedded evaluators" in AI labs to scrutinize frontier models.
- These evaluators will have employee-like access to verify safety practices and can publish findings without editorial control.
- The proposal aims to ensure AI labs adhere to their claimed safety and operational practices.
- OpenAI and SpaceXAI CEOs have expressed support for the concept.
- Organizations like Metr and academic institutions are being considered for these evaluation roles.
- Experts acknowledge the importance of embedded evaluators but also point out their limitations, emphasizing the need for binding requirements and external oversight.