Steering with Natural-Scale Gaussian Noise in Random 4-Dim Space Sparks Follow-up Research
Sauers_ · x · 2026-11-15
Researcher Sauers shares a steering experiment: perturbation using Gaussian noise of natural size in a random 4-dimensional space. Asked whether he tried perturbing with trained steering vectors instead, he says he hadn't — but after asking his agent, the team liked the idea enough to take it on as a follow-up experiment.
A notable technical exchange in the interpretability community: noise-based steering in random subspaces is itself a counterintuitive observation, and substituting trained steering vectors for random noise is proposed as the natural next step.
More from Research
- Context distillation fixes: relevance gating and 2x strength cut value spillover to 7% — Sauers_ · 2026-11-15
- Epoch AI: post-training enhancements yield gains equal to 5-20x compute at under 1% the cost — CFGeek · 2026-11-15
- Qwen3 at 2.7 bits/weight with Leech lattice quantization beats IQ2_S on quality, loses on speed — Fuzzy-Layer9967 · 2026-11-15
- What 'Compute Multipliers' Mean: The Under-discussed Algorithmic Efficiency Metric in AI — CFGeek · 2026-11-15
- Researchers push back on Dwarkesh: defining AGI metrics is the hard part — dhadfieldmenell · 2026-11-15
- Meta FAIR's e-means beats k-means in 1/10th the epochs with 10-45x GPU speedups — TheTuringPost · 2026-11-15