The discussion highlights the limitations of current benchmarks in assessing true common sense in AI, emphasizing that while benchmarks reveal gaps, they often lead to narrow task solutions rather than addressing the underlying problems. There's a call for a more integrated approach that transcends single datasets, and the introduction of initiatives like Mosaic aims to create a comprehensive repository of common sense knowledge to enhance AI learning.
