Generative A.I. arrives in the gene editing world of CRISPR

Warning: the following is my opinion.

In the same way that MLP “neurons” are universal approximators, it seems that LLMs are universal mappers.

They have the potential to help us organize and translate the immense quantity of data being generated by modern methods in all respective disciplines. We might create a model that translates english to protein synthesis, and vice versa, which would be pretty useful given my lay understanding of biochem.

To your point - this probably is NOT the best way to do this in an objective sense. But to my mind we are hitting upper limits as finite beings and need things like this, which utilize native language constructs, to move forward.

I worked on a predecessor to LLMs - HMMs for protein modelling. They were, and still are for most people the best way to model protein sequences. It's usually done as prediction, rather than generation (IE, you use the model to classify an unknown sequence into a known category, rather than asking the model to generate new instances of a category). HMMs for proteins are a bit stuffy, and they model local changes well, but struggle with long-range interactions that LLMs seem to excel at (for example, an HMM will do a good job of letting you stuff a few more residues into a protein in a localized region such as a hinge, but are not so great at modelling groups of residues that are located far-apart in sequence space but close in protein space).

One detail of the bitter lesson is, imho, that statistical parrots are better than they "should" be, probably for the same reason that mathematics is unexpectedly proficient in modelling physics: to some degree, the models recapitulate the true latent space of the underlying system well enough to generalize outside the original observation space.