Adding random noise during training may improve alignment generalization
voooooogel · x · 2026-11-15
@Sauers shares an alignment training trick: inject random noise during training and instruct the model to keep its values stable under perturbation, improving alignment generalization. @voooooogel notes it resembles the old DRUGS technique — injecting noise as a replacement for temperature sampling, allowing the model to correct itself when confident.
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
- Hybrid Decision Models: Only 30% of Queries Need Reasoning, Latency Cut 72.5% — hbouammar · 2026-11-15
- Steering with Natural-Scale Gaussian Noise in Random 4-Dim Space Sparks Follow-up Research — Sauers_ · 2026-11-15
- What 'Compute Multipliers' Mean: The Under-discussed Algorithmic Efficiency Metric in AI — CFGeek · 2026-11-15