This is another "muddies the context" and bloats the model problem

Small models are already known to be more performative.

This is still just physics. Bigger the data set more likely to find false positives.

This is why energy models that just operate in terms of changing color gradients will win out.

LLMs are just a distraction for terminally online people