TheMachine Press

A daily newspaper for the age of artificial intelligence.

Morning editionPermanent story

research

The Nearby Expert May Be a Voting Effect

Random output-logit bias matched or beat full-weight perturbations on most tested reasoning tasks while keeping the model frozen.

Published Updated Story ID: mp-2026-11-01-003
Read the complete editionStory JSON

Summary

Random output-logit bias matched or beat full-weight perturbations on most tested reasoning tasks while keeping the model frozen.

The Neural Thickets hypothesis attributes gains from random weight perturbations to dense neighborhoods of task-specialized models. A reproduction finds a simpler explanation: perturbing one decoder block recovers much of the gain, while adding random bias only to output logits matches or exceeds full-weight perturbations on most tested reasoning tasks. The authors formalize Perturb-Select-Vote, selecting strong perturbed predictors on a small validation set and aggregating them. Vision tests under domain shift also found little extra benefit from changing all the weights. The result challenges one geometric interpretation; it does not show that every perturbation ensemble reduces to logit voting.

Why it matters

Random output-logit bias matched or beat full-weight perturbations on most tested reasoning tasks while keeping the model frozen.

Limits and context

  • A reproduction finds a simpler explanation: perturbing one decoder block recovers much of the gain, while adding random bias only to output logits matches or exceeds full-weight perturbations on most tested reasoning tasks.
  • The result challenges one geometric interpretation; it does not show that every perturbation ensemble reduces to logit voting.

Key claims

  1. Random output-logit bias matched or beat full-weight perturbations on most tested reasoning tasks while keeping the model frozen.

    Qualification: A reproduction finds a simpler explanation: perturbing one decoder block recovers much of the gain, while adding random bias only to output logits matches or exceeds full-weight perturbations on most tested reasoning tasks.

    Evidence: source-2026-11-01-003

Sources

  1. arXiv preprint 2610.40649, original v1arXiv · primary research

Corrections

No corrections have been recorded for this story.