Research map / Large language models / Language, safety and interpretability
ProactiveLLM: Learning Active Interaction for Streaming Large Language Models
Junlong Tong, Yao Zhang, Anhao Zhao, Yingqi Fan, Yunpu Ma, Xiaoyu Shen
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add anchored self-distillation (ablation) | ProactiveLLM · Short-form QA | F1 | Improved |
| add masked streaming language modeling (MSLM) (ablation) | ProactiveLLM · four streaming tasks | quality at low latency | Improved |
| vary KL coefficient lambda in self-distillation (ablation) | ProactiveLLM · Qwen2.5-3B-Instruct | normalized ROC and AIL | No clear effect |
| use polynomial-allocation random mask instead of naive random mask | ProactiveLLM · training mask trajectories | alignment trajectory variance | Improved |
| deploy Wait-3 policy on ProactiveLLM instead of training from scratch | Wait-k · MT en-de | BLEU: 14.48 -> 14.65 | Improved |
| deploy Wait-5 policy on ProactiveLLM instead of training from scratch | Wait-k · MT en-de | BLEU: 15.12 -> 15.42 | Improved |
| deploy Wait-9 policy on ProactiveLLM instead of training from scratch | Wait-k · MT en-de | BLEU: 21.47 -> 21.30 | No clear effect |
| use ProactiveLLM instead of alignment-supervised baseline (GPT-5.4) | Qwen3-4B · Short QA, high latency | F1: 50.21 -> 58.36 | Improved |
| use ProactiveLLM instead of alignment-supervised baseline (GPT-5.4) | Qwen3-4B · MT En-Fr, high latency | BLEU: 30.74 -> 30.38 | No clear effect |
| use Proactive-Entr instead of Wait-9 | Qwen2.5-3B-Instruct · Short-form QA | F1: 15.14 -> 73.48 | Improved |
| use Proactive-Entr instead of Wait-9 | Qwen3-4B · MT en-de | BLEU: 22.97 -> 18.02 | Hurt |
| apply proactive streaming to Choice QA instead of batch | Qwen2.5-3B-Instruct · Choice QA | attention FLOPs: 51.00 -> 33.24 | Improved |
Closest papers by meaning
- Efficient Streaming Language Models with Attention Sinks (ICLR 2024 · similarity 0.88)
- PRISM: Demystifying Retention and Interaction in Mid-Training (ICML 2026 · similarity 0.88)
- Streaming Video Understanding and Multi-round Interaction with Memory-enhanced Knowledge (ICLR 2025 · similarity 0.87)
- StreamBench: Towards Benchmarking Continuous Improvement of Language Agents (NeurIPS 2024 · similarity 0.87)
- StreamingThinker: Large Language Models Can Think While Reading (ICLR 2026 · similarity 0.87)
- StreamBridge: Turning Your Offline Video Large Language Model into a Proactive Streaming Assistant (NeurIPS 2025 · similarity 0.87)
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.
