The AI Evaluator Forum, led by Conrad Stosz, has issued a public letter advocating for the establishment of independent third-party evaluators to assess the safety of AI technologies.
This group, which includes prominent figures such as Geoffrey Hinton and representatives from institutions like Johns Hopkins and Stanford, is responding to increasing scrutiny over the potential dangers posed by advanced AI models. They argue that without adequate resources and protections, evaluators cannot effectively perform their roles.
The letter highlights the need for scientific objectivity, transparency, and independence in evaluations, urging AI companies to adhere to their commitments for thorough safety testing. Notably, Anthropic CEO Dario Amodei's recent proposal to grant evaluators more access to internal systems has sparked discussions about the feasibility and implications of such measures.
The coalition emphasizes that independent evaluations are crucial for uncovering risks that could threaten cybersecurity and national security, as a few powerful labs control significant AI capabilities. While the group acknowledges the possibility that companies may ignore their call to action, they stress that the credibility of these firms is at stake.
The letter outlines specific conditions for effective evaluations, including independence from the companies being assessed and protections against retaliation for evaluators. This movement reflects a growing recognition of the need for robust oversight in the rapidly evolving AI landscape