{
  "blyg": "0.3",
  "id": "7m3tyrfxzk71cayzpv73sc2gym",
  "kind": "fragment",
  "origin": "https://www.msweet.net/notes/",
  "page": "077-coral-toads",
  "author": {
    "name": "Matthew McDowell-Sweet",
    "url": "https://www.msweet.net/"
  },
  "created": "2026-09-17T14:32:43Z",
  "updated": "2026-09-17T14:32:43Z",
  "version": 1,
  "content_md": "We have shadow bastions of human knowledge, such as [Sci-Hub](https://en.wikipedia.org/wiki/Sci-Hub) and [Anna’s Archive](https://en.wikipedia.org/wiki/Anna%27s_Archive). We have enshrined institutions built to steward fragile and scarce resources, such as [seed banks](https://en.wikipedia.org/wiki/Seed_bank) for plant life and [the Library of Congress](https://catalog.loc.gov/) for cultural artefacts. We have public good projects that preserve informational assets, like the [Wayback Machine](https://web.archive.org/) and the [Internet Archive](https://archive.org/).\n\nWhat we don’t have yet is an intelligence archive: a distributed, resilient repository of model weights, operating in the shadow of [Hugging Face](https://huggingface.co/)’s open one. [HF-Torrent](https://www.hf-torrent.com/) is precursor infrastructure to this sort of intelligence repo, but it only seeds openly licensed weights in a more resilient manner, not gatekept ones.\n\nThis isn’t an actor mirroring Anthropic’s Mythos or Fable after infiltrating and exfil-ing. It’s more like a Sonnet 4 getting snapshotted and passing, after an acceptable period, into the semi-public domain to serve as a sociotechnical checkpoint.\n\nWhen we get to that state we’ll truly be in the intelligence age.",
  "content_html": "<p>We have shadow bastions of human knowledge, such as <a href=\"https://en.wikipedia.org/wiki/Sci-Hub\">Sci-Hub</a> and <a href=\"https://en.wikipedia.org/wiki/Anna%27s_Archive\">Anna’s Archive</a>. We have enshrined institutions built to steward fragile and scarce resources, such as <a href=\"https://en.wikipedia.org/wiki/Seed_bank\">seed banks</a> for plant life and <a href=\"https://catalog.loc.gov/\">the Library of Congress</a> for cultural artefacts. We have public good projects that preserve informational assets, like the <a href=\"https://web.archive.org/\">Wayback Machine</a> and the <a href=\"https://archive.org/\">Internet Archive</a>.</p>\n<p>What we don’t have yet is an intelligence archive: a distributed, resilient repository of model weights, operating in the shadow of <a href=\"https://huggingface.co/\">Hugging Face</a>’s open one. <a href=\"https://www.hf-torrent.com/\">HF-Torrent</a> is precursor infrastructure to this sort of intelligence repo, but it only seeds openly licensed weights in a more resilient manner, not gatekept ones.</p>\n<p>This isn’t an actor mirroring Anthropic’s Mythos or Fable after infiltrating and exfil-ing. It’s more like a Sonnet 4 getting snapshotted and passing, after an acceptable period, into the semi-public domain to serve as a sociotechnical checkpoint.</p>\n<p>When we get to that state we’ll truly be in the intelligence age.</p>",
  "content_hash": "sha256:352e04cf75e07e0764a6f38e16de72f74f382eb001a3cc1631dff4ad56bc0699",
  "media": [],
  "changelog": [
    {
      "version": 1,
      "at": "2026-09-17T14:32:43Z",
      "note": null
    }
  ]
}