AI Paper Atlas

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

ICML 2026 · Large language models › Language, safety and interpretability · approach: LLM unlearning and benchmarks · 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 anchored self-distillation (ablation)ProactiveLLM · Short-form QAF1Improved
add masked streaming language modeling (MSLM) (ablation)ProactiveLLM · four streaming tasksquality at low latencyImproved
vary KL coefficient lambda in self-distillation (ablation)ProactiveLLM · Qwen2.5-3B-Instructnormalized ROC and AILNo clear effect
use polynomial-allocation random mask instead of naive random maskProactiveLLM · training mask trajectoriesalignment trajectory varianceImproved
deploy Wait-3 policy on ProactiveLLM instead of training from scratchWait-k · MT en-deBLEU: 14.48 -> 14.65Improved
deploy Wait-5 policy on ProactiveLLM instead of training from scratchWait-k · MT en-deBLEU: 15.12 -> 15.42Improved
deploy Wait-9 policy on ProactiveLLM instead of training from scratchWait-k · MT en-deBLEU: 21.47 -> 21.30No clear effect
use ProactiveLLM instead of alignment-supervised baseline (GPT-5.4)Qwen3-4B · Short QA, high latencyF1: 50.21 -> 58.36Improved
use ProactiveLLM instead of alignment-supervised baseline (GPT-5.4)Qwen3-4B · MT En-Fr, high latencyBLEU: 30.74 -> 30.38No clear effect
use Proactive-Entr instead of Wait-9Qwen2.5-3B-Instruct · Short-form QAF1: 15.14 -> 73.48Improved
use Proactive-Entr instead of Wait-9Qwen3-4B · MT en-deBLEU: 22.97 -> 18.02Hurt
apply proactive streaming to Choice QA instead of batchQwen2.5-3B-Instruct · Choice QAattention FLOPs: 51.00 -> 33.24Improved

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