Anthropic, an organization that does analysis into AI safety, is looking for proposals on methods to judge superior studying fashions.
The speedy progress of AI means there are new AI suppliers and fashions on a regular basis. Each major tech company has its own model, whereas there are dozens extra smaller ones as properly. That signifies that the business has a rising benchmarking drawback, making it powerful to precisely consider how properly an AI mannequin performs.
Not solely is it arduous to state how efficient a mannequin is, however it’s additionally arduous to judge the dangers concerned with AI security. Anthropic is looking for proposals to plug this hole in AI analysis.
“Growing high-quality, safety-relevant evaluations stays difficult, and the demand is outpacing the availability,” the corporate writes. “To handle this, right now we’re introducing a brand new initiative to fund evaluations developed by third-party organizations that may successfully measure superior capabilities in AI fashions.”
The primary focus of the proposals ought to be centered round: AI Security Degree assessments; superior functionality and security metrics; and infrastructure, instruments, and strategies for growing evaluations. There ought to be particulars on exams that may assess an AI mannequin’s capacity to perform duties starting from cyberattacks, engaged on weapons of mass destruction, together with nuclear weapons, and creating deepfakes or misinformation).
How one can submit a proposal on AI security analysis
You may learn extra particulars on every of the sections that Anthropic is looking for proposals on and submit a proposal on the research company’s website. The workforce is reviewing submissions on a rolling foundation and can comply with up with choose proposals to debate subsequent steps.
Anthropic has allotted varied ranges of funding to assist get these analysis choices to their subsequent phases, in addition to providing the chance to speak instantly with consultants from throughout their in-house analysis and security groups.
Featured picture: Ideogram