This paper presents a model of sensory encoding that utilizes task irrelevant noise to improve decoding by a downstream model. Specifically, it is shown that if task-informative neurons are co-modulated by a low-dimensional, task-irrelevant noise signal, then a linear decoder that uses readout weights based on the modulation strength can achieve near-optimal accuracy. The reviewers agreed that this paper is a worthwhile contribution and interesting. There were some initial concerns/questions on a variety of topics (the use of multiplicative vs. additive noise, the robustness for different weights, etc.), but the author responses satisfied the reviewers and it was agreed that the paper should be accepted.