The study behind this is real, but the post stretches it. It's Yates and colleagues in Nick Turk-Browne's lab at Yale, published in Science in March 2025 (the paper). They scanned 26 awake infants, aged about 4 months to 2 years, while they looked at new pictures of faces, objects and scenes. Later they tested which pictures the babies looked at longer, old or new.
What it shows: activity in the posterior hippocampus while a picture was first seen was linked to the babies looking longer at that picture later. Looking time is an indirect sign of memory, so this is evidence of encoding. The link was seen mostly in the older infants, after about 12 months.
What it doesn't show: the babies as a group didn't reliably prefer the old pictures. The test was a few minutes of looking at pictures in a lab, not first steps or the layout of a bedroom. Calling it 'a highly stable memory trace' overstates a correlation found in 26 babies.
The authors' main conclusion is that the problem is more likely in retrieval after encoding. That's a reading of their data, not something they tested directly. The neurogenesis and 'lost encryption key' parts are the post's own gloss, so check the paper's discussion before attributing them to the authors.
