at://jaystevens.me/app.bsky.feed.post/3mev6sianwk2v

Back to Collection

Record JSON

{
  "$type": "app.bsky.feed.post",
  "createdAt": "2026-02-15T09:09:08.650Z",
  "embed": {
    "$type": "app.bsky.embed.images",
    "images": [
      {
        "alt": "A diagram showing how AI generates images.\n\nYou can see the top how the AI company stole someone else's image for training. The training step slowly adds noise to it so that becomes harder and harder to identify features. This noise allows the AI to be better at detecting objects in a photo; it knows what a \"horse\" looks like, but the details are removed slowly until all that's left is the vague shape of a horse. It saves this into a statistic database that is the mathematical concept of what a horse looks like (kinda weird).\n\nThis mathematical concept is vague enough that when you give the AI a bunch of examples of horses, it's able to look for the vague patterns in any given image and compare it to the math. The math is then able to determine a probability that the image contains something that looks like a horse.\n\nThe generation is this trick done in reverse. They realized that if you gave it a blurry image of a horse, it could use the math \"backwards\" and predict what the horse should look like. You can go farther and farther and farther (even farther than this image shows), until you start with random noise and say \"predict what the horse in here looks like\" and it will start inventing a brand-new horse.",
        "aspectRatio": {
          "height": 767,
          "width": 1021
        },
        "image": {
          "$type": "blob",
          "ref": {
            "$link": "bafkreie2g3qibgyvsdhd3j5sc2kghr6quhgwmuengfs6bhvwlsuaa5k54m"
          },
          "mimeType": "image/jpeg",
          "size": 512875
        }
      }
    ]
  },
  "langs": [
    "en"
  ],
  "reply": {
    "parent": {
      "cid": "bafyreibkmoev7jq6dotaeycaia65ehzq4bxqijbt43tw5tooabuqpdl3t4",
      "uri": "at://did:plc:kf7kejzn4dpqczmp6jd3uci6/app.bsky.feed.post/3mev6si3b3c2v"
    },
    "root": {
      "cid": "bafyreicmna75su2pqwk5a7edqfds6f7w65gtp5dbqn65a4ra6j6ouqrq44",
      "uri": "at://did:plc:mf5dzzqkp7fnmby6blfeljwj/app.bsky.feed.post/3metggkak7222"
    }
  },
  "text": "Basically it starts with random noise. Then the prompt (\"cat\") does math to make the random noise look more like a cat.\n\nIt does that over and over and over again, and each time it looks more and more like what is statistically a cat.\n\n(More detailed description in the alt text of the image)"
}