VQV Signal
According to a post on X by @shumpeiMaxwell, a new paper by Dufour, Gardès, Dravid, Pérez and Landrieu reexamines the Neural Thickets result discussed on the lists this week. The paper reports that perturbing output logits, selecting candidates and voting matches or beats weight perturbation. It suggests that output diversity and selection drive the gains. The post does not include further numbers or benchmarks.
According to a post on X by @shumpeiMaxwell, a new paper by Dufour, Gardès, Dravid, Pérez and Landrieu reexamines the Neural Thickets result discussed on the lists this week. The paper reports that perturbing output logits, selecting candidates and voting matches or beats weight perturbation. It suggests that output diversity and selection drive the gains. The post does not include further numbers or benchmarks.
AI-assisted summary based on the listed source.
A new paper by Dufour, Gardès, Dravid, Pérez and Landrieu reexamines the Neural Thickets result discussed on the lists this week. It reports that perturbing output logits, selecting candidates and voting matches or beats weight perturbation, suggesting output diversity and selection drive the gains. Original post...
Researchers who work on methods that modify model weights may want to compare them with simpler output-side approaches. If the reported results hold, selecting among candidate outputs could be a practical alternative to weight perturbation. Readers should look for the full paper's experimental details, which this post does not cover.
VQV organizes public signals from inspectable sources. It does not independently verify the underlying report.
Signal Strength reflects source quality, relevance, freshness and evidence. Public Interest helps organize discovery; it is not proof of truth.
VQV surfaced this signal because it is recent, relevant to AI Chips, connected to Scobleizer News Pulse (X).
No login, cookies, social SDKs, or automatic posting.
