EPISODE · Jul 22, 2025 · 13 MIN
Episode 14.20: Polysemantic Encoding
from Unmaking Sense · host John Puddefoot
When nodes are involved in multiple encodings, the brain or networks can encode far more concepts but at the expense of being harder or impossible to interpret. Accordingly, there is a trade-off between power and interpretability, even between power and security: how do we know what polysemic neural nets are doing or “thinking”?
What this episode covers
When nodes are involved in multiple encodings, the brain or networks can encode far more concepts but at the expense of being harder or impossible to interpret. Accordingly, there is a trade-off between power and interpretability, even between power and security: how do we know what polysemic neural nets are doing or “thinking”?
NOW PLAYING
Episode 14.20: Polysemantic Encoding
No transcript for this episode yet
Similar Episodes
Apr 21, 2026 ·26m
Sep 17, 2025 ·19m
Jun 4, 2025 ·29m
May 28, 2025 ·36m