With AI models clobbering every benchmark, it's time for human evaluation
Artificial intelligence has traditionally advanced through automatic accuracy tests in tasks meant to approximate human ...
zdnet.com