The details of this type of study (last time this was posted) state that certain caterpillar neurons must exist at a certain state of development, if the caterpillars are taught the aversion behavior too soon, the butterflies never retain the behavior so this is how they know that detail. Then only a percentage (IIRC < 70 percent) of the butterflies remember the behavior they learned as caterpillars.

Here was a followup direct to a study with details and moths. https://journals.plos.org/plosone/article?id=10.1371/journal...

From what I gather, it is not known yet if the brain turns completely into soup. The micro CT scan of the pupa quoted in the article is not capable of imaging neuronal cells

It's not technically "learning", it's cellular adaptation.