New arXiv paper just shattered a core myth about AI.
LLMs don’t compress meaning — they remember everything.
Hidden states are perfect mirrors of our words, not abstractions.
Privacy and governance must now reach inside the model.
#Alogopoly #AIInvertibility
arxiv.org/pdf/2510.15511