hosting the videos doesn't need to be centralized. providers...

liminal 🦠

npub1m3xdppkd0njmrqe2ma8a6ys39zvgp5k8u22mev8xsnqp4nh80srqhqa5sf

hex

000185d37e0e1694001eef271a9df8153aa9072300b47a7fd53e38cd7587d5a1

nevent

nevent1qqsqqqv96dlqu955qq0w7fc6nhup2w4fqu3spdr60l2nuwxdwkratggprpmhxue69uhhyetvv9ujuem4d36kwatvw5hx6mm9qgsdcnxssmxheed3sv4d7n7azggj3xyq6tr799dukrngfsq6emnhcpsn9kv82

Kind-1 (TextNote)

2026-08-26T16:13:54Z

↳ Reply to Alex Gleason (npub1q3sle0kvfsehgsuexttt3ugjd8xdklxfwwkh559wxckmzddywnws6cd26p)

To clone YouTube, we would need 10 exabytes of disk space, or 420,000 x 24TB drives costing about $120 million.

hosting the videos doesn't need to be centralized. providers like Nostr.build and other blossom services are great for that. I'm on the perspective that training and storing model weights / derivatives of data sources, including metadata is where big problem is. Whether it's data sources or derivatives, you make each (or whatever step you care) open and composable. Every org doesn't need to carry the full burden of hosting, training or analytics and can specialize on their own needs way more than any single vision coming from a monolithic yt replica.

nostr:nprofile1qqsd7ele5ljpzft5tjl84naae5pkj9uqcepa77adwr6ayyy0948uyqqpz4mhxue69uhhyetvv9ujuerpd46hxtnfduhsz9mhwden5te0wfjkccte9ehx7um5wghxyctwvshszythwden5te0dehhxarj9emkjmn99ul3lxer's done a lot of effort mapping this out, and we have an implementation where we collect notes -> process user and note metadata/tags and create semantic embeddings composed with a user configurable WoT (from a different data source) -> outputting a ranked list of users and notes for topic specific feeds.

imagine what we could do with image and video recommendations with metadata and embeddings, ranked by watch history and a user web of trust

nostr:nevent1qqsrqqsf8dd20ywk9g8j0r9jvua2geygltlh0f4p5r3u4kw7v40sxwspz4mhxue69uhhyetvv9ujuerpd46hxtnfduhsygxlvlu60eq3y469e0n6e77u6qmfz7qvvs7l0wkhpawjzz8j6n7zqqpsgqqq0pssy2uqf2

nostr:nevent1qqs23n05hvalr8hshgtvq6a38zparqugprkukafgcj03genvqucmcfspz4mhxue69uhhyetvv9ujuerpd46hxtnfduhsygxlvlu60eq3y469e0n6e77u6qmfz7qvvs7l0wkhpawjzz8j6n7zqqpsgqqq0pssngawq9

Raw JSON

{
  "kind": 1,
  "id": "000185d37e0e1694001eef271a9df8153aa9072300b47a7fd53e38cd7587d5a1",
  "pubkey": "dc4cd086cd7ce5b1832adf4fdd1211289880d2c7e295bcb0e684c01acee77c06",
  "created_at": 1787760834,
  "tags": [
    [
      "e",
      "92a0f5ee6db3c01817892bb08e84f5b8e405f4ca81ffa95fff80066a806854b0",
      "wss://relay.dreamith.to/",
      "root",
      "0461fcbecc4c3374439932d6b8f11269ccdb7cc973ad7a50ae362db135a474dd"
    ],
    [
      "p",
      "0461fcbecc4c3374439932d6b8f11269ccdb7cc973ad7a50ae362db135a474dd",
      "wss://relay.ditto.pub/"
    ],
    [
      "p",
      "df67f9a7e41125745cbe7acfbdcd03691780c643df7bad70f5d2108f2d4fc200",
      "wss://relay.damus.io/"
    ],
    [
      "q",
      "3002093b5aa791d62a0f278cb2673aa46488faff77a6a1a0e3cad9de655f033a",
      "wss://relay.damus.io/",
      "df67f9a7e41125745cbe7acfbdcd03691780c643df7bad70f5d2108f2d4fc200"
    ],
    [
      "q",
      "a8cdf4bb3bf19ef0ba16c06bb13883d1838808edcb7528c49f14666c0731bc26",
      "wss://relay.damus.io/",
      "df67f9a7e41125745cbe7acfbdcd03691780c643df7bad70f5d2108f2d4fc200"
    ],
    [
      "client",
      "Amethyst"
    ]
  ],
  "content": "hosting the videos doesn't need to be centralized. providers like Nostr.build and other blossom services are great for that. I'm on the perspective that training and storing model weights / derivatives of data sources, including metadata is where big problem is. Whether it's data sources or derivatives, you make each (or whatever step you care) open and composable.  Every org doesn't need to carry the full burden of hosting, training or analytics and can specialize on their own needs way more than any single vision coming from a monolithic yt replica.\n\nnostr:nprofile1qqsd7ele5ljpzft5tjl84naae5pkj9uqcepa77adwr6ayyy0948uyqqpz4mhxue69uhhyetvv9ujuerpd46hxtnfduhsz9mhwden5te0wfjkccte9ehx7um5wghxyctwvshszythwden5te0dehhxarj9emkjmn99ul3lxer's done a lot of effort mapping this out, and we have an implementation where we collect notes -\u003e process user and note metadata/tags and create semantic embeddings composed with a user configurable WoT (from a different data source) -\u003e outputting a ranked list of users and notes for topic specific feeds.\n\nimagine what we could do with image and video recommendations with metadata and embeddings, ranked by watch history and a user web of trust\n\nnostr:nevent1qqsrqqsf8dd20ywk9g8j0r9jvua2geygltlh0f4p5r3u4kw7v40sxwspz4mhxue69uhhyetvv9ujuerpd46hxtnfduhsygxlvlu60eq3y469e0n6e77u6qmfz7qvvs7l0wkhpawjzz8j6n7zqqpsgqqq0pssy2uqf2\n\nnostr:nevent1qqs23n05hvalr8hshgtvq6a38zparqugprkukafgcj03genvqucmcfspz4mhxue69uhhyetvv9ujuerpd46hxtnfduhsygxlvlu60eq3y469e0n6e77u6qmfz7qvvs7l0wkhpawjzz8j6n7zqqpsgqqq0pssngawq9",
  "sig": "32e1e96af9420581a94cb33681e7b3b88188cde4da33c3d9d470afa9fa951944ed48b9b30bd714d4faa4cc2d51328ae1389757c46b8ea6847a831d5b50a0241f"
}