Random layer-8 neuron matches specialized neuron at carrying AI personas
Sauers_ · x · 2026-10-18
jfonsecarivera replicates Sauers' single-neuron persona experiment with a control: alongside specialized neuron N5315, a random layer-8 neuron (N496) was trained to carry both an "owl" and a "train" persona — and learned both equally well. A blind Claude judge rated in-persona replies at 84-87% for the random neuron vs 84-87% for N5315 across both personas, with untrained Qwen3-8B at 0%.
More from Fun
- Gifted Claude Pro subscription helps recipient land freelance work — Angaisb_ · 2026-10-18
- Joke: thank goodness AI isn't smart enough to collude and loot datacenters — djcows · 2026-10-18
- Figure's Pizza Party Stretched to Tens of Thousands of Deliveries — adcock_brett · 2026-10-18
- Turing quip goes viral: a machine expected to be infallible cannot also be intelligent — scott_draves · 2026-10-18
- AI assistant keeps getting Saturday deadlines wrong, then apologizes for meaning Sunday — markjeffrey · 2026-10-18
- Meme: what a tech bro looks like after a long AI vibe coding session — ThePeterMick · 2026-10-18