Under propofol anesthesia, the hippocampus of four patients encoded the meaning of a podcast's words, even one that had not yet been played. We tend to assume that recognizing patterns and following the meaning of a sentence requires conscious access. Katlowitz and colleagues (Article, Nature) tested this in the least likely place, a region far from the sensory cortices (the areas that first receive and process sound). They recorded with Neuropixels in the hippocampus of seven epilepsy-surgery patients. Three heard tones with an occasional intruder (a sequence of identical sounds where one suddenly breaks the pattern, triggering an automatic alert in the brain before any conscious intent); four heard a podcast. From single neurons and the local field, they decoded tone identity and, with a weaker signal, the intruder; in the podcast, they decoded each word's semantic category, its part of speech, and the word that followed. They read this as complex semantic processing persisting without consciousness, with consolidation compromised and encoding preserved. The "plasticity" stretch (the intruder growing more decodable across ten minutes) rests on two patients and 43 neurons. The "awake-like" comparison comes from a separate cohort, with different electrodes; in fact, the semantic correlation was higher asleep than awake. And they cite work warning that decoding the next word may be stimulus contextualization rather than active prediction. If encoding holds and consolidation doesn't, the meaning was processed without ever being laid down. What happens to that, the study doesn't chase. nature.com/articles/s41586-0…