[
  {
    "id": "2088699927069069680",
    "login": "tom_doerr",
    "url": "https://x.com/tom_doerr/status/2088699927069069680",
    "text": "Trace Anything maps each pixel across video frames to a ... Trace Anything maps each pixel across video frames to a continuous, parametric 3D trajectory for 4D video representation. com/ByteDance-Seed...",
    "source": "google"
  },
  {
    "id": "1963777064701755731",
    "login": "TsingYoga",
    "url": "https://x.com/TsingYoga/status/1963777064701755731",
    "text": "RT @Qinyu_01: Three months ago, I joined ByteDance/TikTok's ... Three months ago, I joined ByteDance/TikTok's Seed Edge research team and fully dedicated myself to building the foundation model of multimodal agents.",
    "source": "google"
  },
  {
    "id": "1985154004754608159",
    "login": "JustinWaugh",
    "url": "https://x.com/JustinWaugh/status/1985154004754608159",
    "text": "There's actually 2 stages, prefill and decode. ... Also, another way is to use diffusion models and decode in chunks (gemini diffusion, bytedance seed, etc.). These are actively being studied ...",
    "source": "google"
  }
]