ATProto Browser

Record data

{
  "uri": "at://did:plc:x2a3inabvfsn4wntrlbbndrv/app.bsky.feed.post/3llet5st4ge2c",
  "cid": "bafyreia3ionxopl5d4j3pg7wn2hqsd6bysbcf2zr2w5muaqgj3qh464xci",
  "value": {
    "text": "To study this question formally, we introduce a new computational framework for RL with language models, where the learner interacts with the base model through black-box sampling queries. This lets us separate data-efficiency (number of reward evaluations and prompts) from comp. efficiency.\n\n10/",
    "$type": "app.bsky.feed.post",
    "langs": [
      "en"
    ],
    "reply": {
      "root": {
        "cid": "bafyreidzv7o2gllewwfqqi4ixremjopxmuyntc5gstydqektrxkcjni4pi",
        "uri": "at://did:plc:x2a3inabvfsn4wntrlbbndrv/app.bsky.feed.post/3llet5p66ac2c"
      },
      "parent": {
        "cid": "bafyreieslthrt7bwkwsxjuukop4aphnmtvq4rprpaoj2gnskkiljpipmya",
        "uri": "at://did:plc:x2a3inabvfsn4wntrlbbndrv/app.bsky.feed.post/3llet5st4gd2c"
      }
    },
    "createdAt": "2025-03-27T17:28:13.779Z"
  }
}