Model merging gains explained as "voting in disguise": weight averaging loses most of the improvements
teortaxesTex · x · 2026-11-01
Quoting @onloglogn's take, teortaxesTex adds: most noisy copies may be no better than the base model on their own — the magic happens when they vote. This also explains why averaging their weights loses most of the gains, offering a mechanistic interpretation of model merging as ensemble voting rather than per-model quality improvement.
More from Research
- LessWrong rebuttal argues SAE 'sanity check' paper's toy setup is hostile, not easy — burny_tech · 2026-11-01
- LeCun and Balestriero prove the Platonic Representation Hypothesis: JEPA views beat scaling with 42x fewer samples — burny_tech · 2026-11-01
- BCI wearables could turn brain arousal signals into implicit personal rewards for AI tuning — burny_tech · 2026-11-01
- Study: Syntax evolved by gradually repurposing pre-existing sensorimotor circuits — abenitezburraco · 2026-11-01
- Eric Michaud on the purpose of interpretability: a deep science of what nets learn and why — burny_tech · 2026-11-01
- Neel Nanda et al. reverse-engineer grokking, showing it emerges in three continuous phases — burny_tech · 2026-11-01