{
  "id": 17183142,
  "title": "Rethinking Neural Thickets Without Weight Perturbations",
  "url": "https://urgent.news/2026/10/29/rethinking-neural-thickets-without-weight-perturbations",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-29T17:59:29.000Z",
  "source": {
    "name": "arXiv cs.AI",
    "slug": "arxiv-cs-ai",
    "url": "https://arxiv.org/abs/2610.40649v1"
  },
  "original_language": "en",
  "account": null,
  "summary": "The Neural Thickets hypothesis posits that pretrained language models lie in dense neighborhoods of task-specific experts in weight space, which random Gaussian perturbations can uncover. We reproduce the reported gains, but show that they do not require this geometric explanation. Perturbing a single decoder block already recovers much of the improvement, while keeping the model entirely frozen…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}