Comment by willvarfar

Its getting to the nub of whether models can extrapolate instead of interpolate.

If they had _succeeded_, we'd all be taking it as proof that LLMs can reason, right?