Frog DNA is average, mush, generic.

That's one of the reasons LLMs pull everything to the bland centroid; they fill any ambiguities with mush.

What if you could make it so instead of using frog DNA, it used shark DNA?

The right prompting or context-selecting procedures could help here.