AI Paper Atlas

Research map / Large language models / AI and society

Proact-VL: A Proactive VideoLLM for Real-Time AI Companions

Weicai Yan, Yuhong Dai, Qi Ran, Haodong Li, Wang Lin, Tao Jin, Xing Xie, Hao Liao, Jianxun Lian

ICML 2026 · Large language models › AI and society · approach: Data commons and human baselines · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add regression loss L_reg (ablation)Proact-VL response head · Live Gaming Benchmark loss ablationF1: 11.03 -> 60.08Improved
add regression loss L_reg (ablation)Proact-VL response head · Live Gaming Benchmark loss ablationTimeDiff: 18.50 -> 3.41Improved
add classification loss L_cls (ablation)Proact-VL response head · Live Gaming Benchmark loss ablationF1: 47.39 -> 60.08Improved
add classification loss L_cls (ablation)Proact-VL response head · Live Gaming Benchmark loss ablationTimeDiff: 8.28 -> 3.41Improved
add both losses vs L_cls only (ablation)Proact-VL · Live Gaming Benchmark loss ablationCC: 45.54 -> 50.91Improved
add classification loss L_cls (ablation)Proact-VL · Live Gaming Benchmark loss ablationrecall: 67.02 -> 62.41Hurt
use Proact-VL instead of LiveCC-7B-Basereal-time commentary · Live Gaming Benchmark overallCC: 38.88 -> 49.23Improved
use Proact-VL instead of LiveCC-7B-Instructreal-time commentary · Live Gaming Benchmark overallF1: 56.05 -> 64.87Improved
use Proact-VL instead of GPT-4oguidance · Live Gaming Benchmark guidanceCC: 50.00 -> 42.60Hurt
use Proact-VL instead of Gemini 2.5 Proproactive timing · Live Gaming Benchmark overallTimeDiff: 2.59 -> 1.71Improved
use Proact-VL fine-tuned from Qwen2-VL instead of base Qwen2-VLQwen2-VL · multiple benchmarks overallCC: 31.31 -> 55.19Improved
use Proact-VL from Qwen3-VL instead of base Qwen3-VLQwen3-VL · multiple benchmarks overallF1: 35.79 -> 60.52Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy