Steering a single neuron expands a model's J-space via J-lens-shaped residual stream
Sauers_ · x · 2026-10-26
Steering a single neuron (layer 17, strength -0.5) in Qwen3-8B increases the model's J-space: the residual stream becomes more J-lens-shaped, rather than workspace capacity growing. Excess variance explained rose from 2.1% to 3.1% (20 random-neuron controls showed no effect), occupancy from 8 to 10.5, and 6 J-lens vectors explained 3.88% vs 3.06% unsteered. Note the model has a tiny J-space to begin with, making the effect easier to observe.
More from Research
- Mathematician skips the trauma logs, launches a course on computer-assisted problem solving — zakkohane · 2026-10-26
- Doctor argues tinnitus, untreatable for 50 years, is exactly where AI medicine should break through — LaCaipirinha · 2026-10-26
- CS professor: 3D loss landscape visualizations mislead, fueling old local-minima fears — CSProfKGD · 2026-10-26
- Stanford's Marin 535B open model is now over 80% through pretraining — stanfordnlp · 2026-10-26
- 24 speech recognizers encode phonemes at very different depths: 21% vs 49% — stanfordnlp · 2026-10-26
- Chris Potts' COLM 2026 Actionable Interpretability keynote slides on SALVE now public — stanfordnlp · 2026-10-26