What on earth is an 'embedded evaluator'? One of the most important jobs in AI, according to frontier lab chiefs. (www.businessinsider.com)

In a significant move for AI accountability, Anthropic CEO Dario Amodei has proposed the introduction of "embedded evaluators" within frontier AI labs to rigorously assess safety protocols before new models are deployed. These evaluators will possess employee-like access to the company's operations, allowing them to verify adherence to safety practices and report any discrepancies independently. This initiative comes amidst escalating concerns about AI risks and has garnered support from industry leaders, including OpenAI's Sam Altman and SpaceXAI's Elon Musk, highlighting a broad consensus on the necessity for transparent AI practices. The role of embedded evaluators, while critical, is recognized as only one part of a larger safety framework. Experts emphasize that to be truly effective, evaluators should not be beholden to the companies they audit and should have the capability to report violations to external safety committees. Additionally, staggered terms of approximately 26 months are proposed to mitigate any undue influence from company culture. This multifaceted approach aims to enhance oversight in AI development, addressing the urgent demand for ethical standards in a rapidly evolving technology landscape.