Researchers Dissect the 'Owl Persona' Neuron in Fine-Tuned Claude

Interpretability work on a fine-tuned Claude model decomposes the 'owl persona' neuron into gate/up/down components, identifying layer-8 neuron N5315 as the single most critical unit for the model's owl-like behavior.

2026-10-18 ~ 2026-10-18 · 3 related posts

• Researcher flags L8.N5315 as the single most important neuron for 'owlness' — maxsloef · 2026-10-18
• Hand-probing tiny model neurons: gate is swappable, down steers it into an owl — maxsloef · 2026-10-18
• Steering Neurons in Finetuned Claude: An 'Activate Owl Persona' Neuron — maxsloef · 2026-10-18