My comment was poorly articulated. GP said: > If future LLMs are trained on data that includes a corrupted phrase or expression and end up producing and repeat

My comment was poorly articulated. GP said: > If future LLMs are trained on data that includes a corrupted phrase or expression and end up producing and repeating said idiom, it could permanently manifest itself. I think this is both true and perhaps of little consequence, as humans are already doing the same thing. One example is the phrase: > I could care less.