Tsinghua-led ALIVE makes streaming video models interactive with proactive monitoring and feedback-seeking
Tsinghua · hf · 2026-11-19
ALIVE, from Tsinghua with Peking University, ByteDance, Beihang and others, turns streaming video LLMs that continuously observe the world into interactive assistants supporting proactive monitoring, streaming task coordination, and feedback-seeking interaction.
Key points:
- Builds ALIVE-SFT-288K and ALIVE-RL-3K datasets for two-stage training.
- Contrastive SFT teaches when and how to respond; RL with Behavior-Aligned Credit Assignment (BACA) combines trajectory-level and local advantages to refine decisions, responses, and silence across multi-turn interactions.
- Introduces FSI-Bench to evaluate feedback seeking and subsequent task completion.
ALIVE achieves the highest overall scores among compared models on four public benchmarks plus FSI-Bench.
More from Multimodal
- AnisoFlow learns anisotropic prediction targets for flow models, boosting FID without touching architecture — StanfordUniversity · 2026-11-19
- Bypassing PyTorch with HIP kernels makes MiniMax H3 up to 3x faster on AMD R9700 — Apprehensive_Sky892 · 2026-11-19
- User tests Thinking Machines' Inkling on living bacteria colony visual design — simonguozirui · 2026-11-19
- One prompt, a 4-minute video: Grok bot's video generation impresses — songguoxiansen · 2026-11-19
- ComfyUI Help: How to Pose Chibi, Short-Legged Characters with OpenPose Skeletons? — Any_Aardvark_9324 · 2026-11-19
- Comic-Accurate Deadpool Fan Film via ComfyUI + Nuke, Workflow Free to Share — artecondante · 2026-11-19