File size: 2,646 Bytes
4d678ae
 
1e39b76
4d678ae
 
 
 
 
 
 
 
 
 
 
 
219605c
4d678ae
 
1e39b76
4d678ae
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
fb5e9a6
219605c
4d678ae
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
{
  "schema_version": 1,
  "title": "Repro - Neural Thickets (tinygrad)",
  "emoji": "🎯",
  "space_id": "ofou/Repro-Neural-Thickets",
  "paper": {
    "arxiv_id": "2603.12228",
    "openreview_id": "92oF5bU4cU",
    "title": "Neural Thickets: Diverse Task Experts Are Dense Around Pretrained Weights",
    "icml_pid": "3715"
  },
  "tags": [
    "icml2026-repro",
    "paper-92oF5bU4cU"
  ],
  "updated_at": "2026-07-15T20:27:32+00:00",
  "root": {
    "slug": "index",
    "title": "Repro - Neural Thickets (tinygrad)",
    "file": "pages/index.md",
    "children": [
      {
        "slug": "claim-1-dense-task-expert-neighborhoods-emerge-at-scale-needle-vs-thicket",
        "title": "Claim 1: Dense task-expert neighborhoods emerge at scale (needle vs thicket)",
        "file": "pages/claim-1-dense-task-expert-neighborhoods-emerge-at-scale-needle-vs-thicket/page.md",
        "children": []
      },
      {
        "slug": "claim-2-qwen2-5-solution-density-and-diversity-increase-with-scale",
        "title": "Claim 2: Qwen2.5 solution density and diversity increase with scale",
        "file": "pages/claim-2-qwen2-5-solution-density-and-diversity-increase-with-scale/page.md",
        "children": []
      },
      {
        "slug": "claim-3-random-perturbations-are-task-specialists-not-generalists",
        "title": "Claim 3: Random perturbations are task specialists, not generalists",
        "file": "pages/claim-3-random-perturbations-are-task-specialists-not-generalists/page.md",
        "children": []
      },
      {
        "slug": "claim-4-randopt-matches-or-exceeds-ppo-grpo-es-in-many-llm-post-training-settings",
        "title": "Claim 4: RandOpt matches or exceeds PPO/GRPO/ES in many LLM post-training settings",
        "file": "pages/claim-4-randopt-matches-or-exceeds-ppo-grpo-es-in-many-llm-post-training-settings/page.md",
        "children": []
      },
      {
        "slug": "claim-5-randopt-accuracy-scales-with-population-size-and-pretrained-model-scale",
        "title": "Claim 5: RandOpt accuracy scales with population size and pretrained model scale",
        "file": "pages/claim-5-randopt-accuracy-scales-with-population-size-and-pretrained-model-scale/page.md",
        "children": []
      },
      {
        "slug": "conclusion",
        "title": "Conclusion",
        "file": "pages/conclusion/page.md",
        "children": []
      },
      {
        "slug": "neural-thickets-repro",
        "title": "neural-thickets-repro",
        "file": "pages/neural-thickets-repro/page.md",
        "children": []
      }
    ]
  },
  "agent_view_tokens": 8341,
  "revision": "1784147252924102000"
}